JP2014154030A - Subject-verb agreement error detection device and program for agreement error detection - Google Patents

Abstract

PROBLEM TO BE SOLVED: To achieve more simple and accurate detection of a subject-verb agreement error in an input sentence in English without using syntax analysis.SOLUTION: A subject-verb agreement error detection device 12 includes: input sentence analysis means 14 which performs part-of-speech analysis and phrase analysis of an input sentence in English, specifies words, phrases and clauses configuring the input sentence, and acquires part-of-speech information regarding the kinds and the number of the parts of speech of the words and phrase and clause information regarding the kinds of the phrases and clauses; subject-verb candidate extraction means 15 which extracts a verb in a verb phrase in the input sentence based on rules stored in advance without performing syntax analysis by using the result of the input sentence analysis means 14, and extracts a noun in a noun phrase existing on the left side of the verb as a subject candidate; and correctness/incorrectness determination means 16 which determines correctness/incorrectness of the agreement between the subject candidate and the verb based on the rules stored in advance from the person and the number of the verb and the subject candidate respectively extracted by the subject-verb candidate extraction means 15.

Description

  The present invention relates to an apparatus and a computer program for detecting a coincidence error in which the subject person / number of the subject does not match the person name / number of the verb for an English sentence created by an English learner, and more specifically, without using parsing. Furthermore, the present invention relates to a matching error detection apparatus and a matching error detection program for a subject verb that can detect the matching error more accurately and easily from only part-of-speech analysis and phrase analysis.

As an error when a Japanese learner of English creates an English sentence, for example,
"The football player from Japan play for United."
As described above, the frequency of errors in which the subject's personality / number does not match the verb's personality / number (hereinafter referred to as “subject verbal matching error”) is relatively high. As an apparatus for detecting such a subject verb matching error, there is a learning support apparatus disclosed in Patent Document 1. In this learning support device, the words corresponding to the subject and the verb are identified from the input sentences in English, the grammatical attribute of each word is acquired from the word dictionary, and whether or not the subject and the verb personality match. Is detected. However, Patent Document 1 does not specifically disclose how to determine the subject and verb of an input sentence.

  By the way, as a method for detecting a matching error of a subject verb, a Chodrow method for detecting the matching error based on a concatenation probability of a few words (or parts of speech) is known (see Non-Patent Document 1). The matching error can be detected by this method only when the subject and the verb are adjacent to each other or when the distance between the subject and the verb is one word. However, as in the case of the above-described example sentence, it is not uncommon for an English sentence to have an interval between a subject and a verb of two words or more. Therefore, the method cannot accurately detect the matching error in such a case. Therefore, in order to accurately detect the matching error, it is necessary to accurately grasp the relationship between the subject in the sentence and the verb. For this purpose, the English sentence created by the English learner is analyzed. A technique described in Non-Patent Document 2 is known in which after a relationship between a subject and a verb is acquired by the syntax analysis, a matching error of the subject verb is detected.

JP-A-8-30598

M. Chodorow and C. Leacock, An unsupervised method for detecting grammatical errors, Proc. Of 1st Meeting of the North America Chapter of ACL, Oct. 2000, PP140-147 Ikuo Kawai, Atsuyoshi Sugihara, Noboru Sugie, “A system for detecting English errors ASPEC-I” IPSJ Transactions Vol.25 No.6, November 1984, pp1072-1079

  However, the method described in Non-Patent Document 2 has the following problems. In other words, parsing is a task that is difficult to be accurate. For example, when the subject is composed of a large number of words, it is necessary to make a judgment considering the meaning and attributes of the words. There is a limit to always accurately identifying the subject and verb. In particular, the English sentence created by the English learner includes not only the matching error of the subject verb but also various other errors, and in such a case, the syntax analysis becomes more difficult. Therefore, even if syntactic analysis is used, it cannot be said that the subject verb matching error can always be detected accurately. In addition, syntax analysis is generally developed using a large amount of English data with syntax information added manually, and it takes a lot of time and money to construct and use the syntax analysis. Therefore, at present, it cannot be said that the subject verb matching error detection method using parsing is useful.

  The present invention has been devised by paying attention to such a problem, and its purpose is to detect a subject verb matching error more easily and accurately in an input sentence in English without using syntax analysis. An object of the present invention is to provide a matching error detection device and a matching error detection program for a subject verb.

To achieve the above object, the present invention mainly provides a subject verb matching error detection device for detecting matching errors between a subject and a verb in an input sentence in English,
Perform part-of-speech analysis and phrase analysis of the input sentence, specify words, phrases and clauses constituting the input sentence, part-of-speech information related to the type and number of parts of speech of the word, and phrase and clause information related to the type of phrase and clause Using the result of the input sentence analyzing means, and extracting the verb in the verb phrase for the input sentence based on the rules stored in advance without performing syntax analysis, and the verb The subject verb candidate extracting means for extracting a noun in the noun phrase existing on the left side of the subject as a subject candidate, and the personality and number of each of the verb and the subject candidate extracted by the subject verb candidate extracting means are stored in advance. In accordance with the rules, correctness determination means for determining whether or not the subject candidate matches the verb is provided.

  In the evaluation experiment conducted by the inventor, it has been proved that the technique applied to the present invention has a higher detection accuracy of the subject verb matching error in the input sentence than the conventional technique using syntax analysis. According to the present invention, it is possible to more easily and accurately detect a subject verb matching error in an input sentence without using parsing.

The block diagram which shows schematic structure of the English learning system provided with the coincidence error detection apparatus of the subject verb which concerns on this invention. The chart figure showing the process sequence of the said coincidence error detection apparatus.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 is a block diagram showing a schematic configuration of an English learning system including a subject verb matching error detection apparatus according to the present invention. In this figure, an English learning system 10 includes an input device 11 to which an English sentence created by a user who is an English learner is input, and an input sentence input by the input device 11 with a subject whose personal name / number is a verb. A subject verb matching error detection device 12 that detects a grammatical error that does not match the number. The following description will be made on the premise of an input sentence consisting of one sentence. If an English sentence consisting of a plurality of sentences is input to the input device 11, a process for cutting out one sentence at a time is performed based on the position of the period and the like, and a process described later is performed for each sentence.

  The input device 11 is constituted by a data input device such as a keyboard or a touch panel (not shown), but is not limited to this, and reads English text recorded on a paper medium as image data. It may be a scanner device that converts text data into text data, or a device that can read text data related to English text stored in a storage medium.

  The coincidence error detection device 12 is constituted by a computer including an arithmetic processing unit such as a CPU, a storage device such as a memory and a hard disk, and the like is installed with a program for causing the computer to function as each unit described later. Here, although not particularly limited, the matching error detection device 12 of the present embodiment is a server that can transmit and receive data to and from many user input devices 11 via a network line such as the Internet line. Is provided.

  The coincidence error detection device 12 uses the input sentence analysis means 14 for analyzing the input sentence from the input device 11 and the result of the input sentence analysis means 14, and does not perform syntax analysis from the input sentence. And subject verb candidate extraction means 15 for extracting subject candidates corresponding to the verb, and information on the names and numbers of the verbs and subject candidates extracted by the subject verb candidate extraction means 15. A correctness / incorrectness determination means 16 for determining correctness / incorrectness related to the match and a database 17 in which various data necessary for processing in each of the means 14 to 16 are stored.

  The input sentence analysis means 14 includes a part-of-speech analysis unit 19 that performs part-of-speech analysis on each word constituting the input sentence, and a phrase analysis that performs phrase analysis on the input sentence using the analysis result in the part-of-speech analysis unit 19 Part 20.

  In the part-of-speech analysis unit 19, each constituent word is extracted due to the presence of a space in the input sentence, and data stored in the database 17, that is, dictionary data such as part-of-speech name and meaning of each word, and part-of-speech of an adjacent word The part-of-speech information of each word is specified from the probability data of the type. That is, here, in addition to the part-of-speech name, a code (part-of-speech label) representing the person name, the number of singular or plural forms, the utilization form, etc. is attached to each word.

  In the phrase analysis unit 20, based on the part-of-speech information of each word obtained by the part-of-speech analysis, each word is obtained from the data stored in the database 17, that is, from the dictionary data and the adjacent part-of-speech and phrase type probability data. Are grouped in units of phrases and clauses in order from the beginning of the input sentence. That is, here, the input sentence is divided into phrase units such as noun phrases, verb phrases, adjective phrases, adverb phrases, conjunction phrases, etc., and clause units such as noun clauses, verb clauses, etc. And a code (phrase label, clause label) indicating the phrase and clause information of the type.

  Note that the processing in the part-of-speech analysis unit 19 and the phrase analysis unit 20 is performed based on a known method and is not an essential part of the invention, and thus detailed description thereof is omitted.

  The subject verb candidate extraction means 15 normalizes the input sentence normalization processing unit 22 for normalizing the input sentence to improve the extraction accuracy of verbs and subject candidates, and the input sentence normalization processing part 22 An extraction processing unit 23 that performs processing for extracting verbs and subject candidates of the input sentence for the input sentence after processing is provided.

  The input sentence normalization processing unit 22 is based on a target selection processing unit 25 that performs a process of selecting a sentence to be detected as a subject verb matching error based on the type of the input sentence, and a noun phrase in which nouns are arranged in parallel. A parallel noun analysis processing unit 26 that performs adjustment processing so that subject candidates can be accurately extracted, an input sentence division processing unit 27 that performs input sentence division processing to narrow the range of verb and subject candidate extraction, and input An unnecessary phrase deletion processing unit 28 that deletes phrases that cannot be the subject of a sentence to reduce unnecessary subject candidates, and an inversion syntax of the subject and verb starting from There, It is composed of an inverted syntax conversion processing unit 29 that performs word order conversion processing for rearranging like a normal syntax.

  The target selection processing unit 25 excludes question sentences and command sentences from the input sentence and extracts only a plain text. In other words, the matching error detection device 12 detects a matching error of the subject verb only for the input sentence that is a plain text. Specifically, when a question mark (?) Is included in the input sentence, the input sentence is determined to be a question sentence and is excluded from detection of a subject verb matching error. Further, when a subject candidate is not obtained by the processing to be described later in the extraction processing unit 23, the input sentence is determined to be a command sentence and is excluded from the subject verb matching error detection target.

In the parallel noun analysis processing unit 26, the following processing is performed. For example,
"Bread and butter are expensive"
Like “bread” and “butter” in a single noun phrase “bread and butter”, each is a singular noun, but since they are arranged in parallel, the noun phrase “bread and butter” As the treatment, the personality / number of the verb located after (right side) in the sentence is specified.
Therefore, in the parallel noun analysis processing unit 26, when nouns are arranged in parallel with “and” or “or” in one noun phrase, a special code meaning “parallel noun” for each part of speech of the words. (Part of speech label) is attached and stored in correspondence with these words. That is, here, the number of subject candidates is adjusted, and the part-of-speech label replacement process is performed to properly detect the subject verb matching error. It is possible to accurately determine singular and plural subject candidates that have influence.

  In the input sentence division processing unit 27, the input sentence is divided at a predetermined position based on the next rule set in advance and stored in the database 17. In the following description, each part divided by the input sentence division processing unit 27 is referred to as a segment. Extraction of subject candidates and verbs by the extraction processing unit 23 is performed for each segment divided by the input sentence division processing unit 27.

  First, the input sentence is divided immediately before a predetermined clause. That is, if a subordinate conjunction clause exists by analysis of the input sentence by the phrase analysis unit 20 or a keyword stored in the database 17 is included in advance, the input is performed immediately before the subordinate conjunction clause and the keyword. The sentence is split. As keywords here, “but”, “if”, “because”, “since”, “though”, “although”, “how”, “what”, “where”, “where”, “where” , “While”.

For example, if the input sentence is
"The football player from Japan play for United when he thinks that it is right."
In this case, the input sentence is divided immediately before “where” of the keyword and immediately before “that” at the head of the subordinate conjunction clause, and three segments are obtained.

  In addition, when it is determined from the analysis result in the part-of-speech analysis unit 19 that there is a related pronoun (“who”, “who”) in the input sentence, among the noun phrases existing before (left side) of the related pronoun The input sentence is divided even immediately before the noun phrase closest to the related pronoun.

  The unnecessary phrase deletion processing unit 28 deletes unnecessary phrases based on the following rules set in advance and stored in the database 17. That is, first, phrases other than the noun phrase and the verb phrase are deleted for each segment from the analysis result of the input sentence in the phrase analysis unit 20. Here, nouns with prepositions do not become the subject, so noun phrases adjacent to the right after the preposition phrase are also deleted in the noun phrases.

For example, if the input sentence is
“The football player with glasses play for Ajax.”
In the phrase analysis unit 20, "The football player" is a noun phrase, "with" is a preposition phrase, "glasses" is a noun phrase, "play" is a verb phrase, "for" is a preposition phrase, and "Ajax" Is determined as a noun phrase. Therefore, in this case, the unnecessary phrase deletion processing unit 28 deletes the preposition phrase “with”, the noun phrase “glasses”, the preposition phrase “for”, and the noun phrase “Ajax”, and the noun phrase “The football player” and the verb The phrase “play” will remain.

  Here, as an exception, noun phrases such as “a number of people” and “most of them” that are adjacent to “noun phrases + of” that act as quantifiers are modified as adjectives. As a rule, no deletion is performed. That is, when “a number of” and “a couple of” exist as the first exception rule, the part of “noun phrase + of” is deleted, and the number information of the noun is handled in plural for the following noun phrase. The code (part-of-speech label) representing the numerical information is replaced. In addition, as the second exception rule, “most of”, “any of”, “some of”, “many of”, “(a) new of”, “(a) small of”, “a lot of” , “(A) friendship of” is deleted, “noun phrase + of” is deleted.

The inversion syntax conversion processing unit 29 performs processing for returning the inversion syntax of the subject and verb using “There” meaning presence to the normal word order, and the extraction processing unit 23 extracts the verb and subject candidate. It is easy to do. Specifically, since “There” meaning existence is found by analysis in the part-of-speech analysis unit 19, the sign (part-of-speech label) is detected, and when there is a verb phrase immediately after “There”, “ From “There” to the next verb phrase of the verb phrase immediately after “There”, or to the end of the sentence when there is no next verb phrase, they are rearranged in reverse order at the phrase level. For example, if the input sentence is
“There is a key.”
In the case of the inversion syntax conversion processing unit 29,
“A key is There.”
Sorted into

  In the extraction processing unit 23, the following verb is set for each segment with respect to the input sentence normalized by the input sentence normalization processing unit 22 based on the next rule set in advance and stored in the database 17. And the subject candidate of the verb.

  First, verbs are extracted by using analysis results obtained by the input sentence analysis means 14, that is, part of speech information and phrase information. Here, the subject verbs are subject to the restriction of the present verb except for the Be verb. First, for verbs other than the Be verb, the verb non-third-person present or verb third-person present part-of-speech label. Extracted for verbs marked with. Here, in the part-of-speech analysis unit 19, the verb with the part-of-speech label of the verb prototype is also extracted in order to cope with the analysis error of the part-of-speech analysis that is analyzed with the verb non-third-person present. Be targeted. However, verb prototypes used with auxiliary verbs such as “can” and “will” are not extracted. In addition, a To infinitive such as “To read” is analyzed as a verb phrase, but “To” here is analyzed as a To infinitive as part of speech information, and “read” is extracted as a verb. It is assumed to be outside.

  On the other hand, for the Be verb, matching errors are detected based on the surface information of the word as will be described later. In the extraction processing unit 23, “am”, “are”, “is”, “ When there are five types of Be verbs “was” and “where”, they are extracted as verbs.

  For the subject candidates, nouns, pronouns, and numbers in the noun phrase in front of the verb extracted by the above processing (left side) are extracted. At this time, To infinitives and verbal nouns in front of the verb (left side) can be the subject, but are analyzed as verb phrases by the phrase analysis unit 20, so that pattern matching based on patterns stored in the database 17 in advance is performed. Processing is performed and extracted as a subject candidate. In this case, for the To infinitive, for convenience, the infinitive “To” is a subject candidate, and in the case of a verbal noun, a word that is a verbal noun is a subject candidate.

  Then, a person name and a number are determined for each extracted subject candidate. Common nouns and proper nouns are all third person. The number information such as singular and plural of the nouns is determined based on the adjustment of the number information of words performed by the input sentence normalization processing unit 22 together with the part of speech information acquired by the part of speech analysis unit 19. Here, among the subject candidates, when the part of speech information acquired by the part of speech analysis unit 19 is a collective noun that can be used as both a singular noun and a plurality of nouns in the original form, it is determined to have a singular and a plurality of two numbers. Is done. In addition, when there are a plurality of subject candidates for one verb, a plurality of pieces of number information are provided regardless of the number information of nouns that are subject candidates. That is, the number information of the noun is determined according to the corresponding rule shown in the following table based on the part of speech label. On the other hand, for pronouns, the person name / number is not determined, and matching errors are detected based on the surface information of the word as will be described later.

  Here, in order to reduce the influence of the spelling error of the input sentence, the verb and the subject candidate may be extracted after correcting the misspelled word correctly by using a spell checker (not shown). good. In other words, the spell checker checks whether there is a spelling error, and if a spelling error is found, an error correction candidate is obtained, and the person / number of the candidate word can be determined as described above.

  The correctness / incorrectness determination means 16 performs correctness / incorrectness determination regarding the matching of subject verbs based on the detection rules in the table below stored in the database 17 in advance. In the table below, “◯” indicates that the subject verbs have the same personality / number, and “x” indicates that the subject verbs do not have the same personality / number. That is, here, it is determined whether or not the person / number of each verb extracted by the extraction processing unit 23 matches the corresponding subject candidate. As a result, it is determined that the matching of the subject verb is incorrect only when the subject names / numbers of all the subject candidates for one verb are incorrect. In other words, no matter which subject candidate is selected as the subject corresponding to the verb, an error is detected only when the person / number does not match.

  Next, the processing procedure of the matching error detection device 12 will be described with reference to FIG.

  First, after an input sentence is input, the input sentence analysis unit 14 performs an input sentence analysis process for performing part-of-speech analysis and phrase analysis of each word constituting the input sentence (step S101).

  Thereafter, the input sentence normalization processing unit 22 performs an input sentence normalization process.

  Specifically, first, in the target selection processing unit 25, when the input sentence is a question sentence, a target exclusion process is performed to exclude it from the subject verb coincidence error detection target (step S102).

Next, the part-of-speech label replacement processing of the word specified by the part-of-speech analysis in the part-of-speech analysis unit 19 is performed on a predetermined noun phrase (step S103). That is, the replacement processing here is performed by the parallel noun analysis processing unit 26. When nouns are arranged in parallel with “and” or “or” in one noun phrase, parallel nouns are respectively obtained for the parts of speech of the words. The part-of-speech label representing In addition, in the unnecessary phrase deletion processing unit 28, when it corresponds to the first exception rule, that is,
When “a number of” and “a couple of” exist, the part of “noun phrase + of” is deleted, and the part of speech label is changed so that the number information of the noun is made plural for the following noun phrase .

  Thereafter, the input sentence division processing unit 27 performs input sentence division processing for dividing the input sentence into segments (step S104). Here, when it cannot be divided in the middle of the input sentence according to the rules described above, it is handled as one segment.

  Then, the unnecessary phrase deletion processing unit 28 performs an unnecessary phrase deletion process in which phrases other than the predetermined noun phrase and verb phrase are deleted (step S105). The phrase deleted here is excluded from the subject of the subject verb matching error detection.

  Next, in the inverted syntax conversion processing unit 29, when there is an inverted syntax of a subject and a verb using “There” meaning presence, a word order conversion process is performed to rearrange them in reverse order at the phrase level (step S106). .

  When the above input sentence normalization processing is completed, the extraction processing section 23 performs subject verb candidate extraction processing for extracting the verb and the corresponding subject candidate for each segment (step S107). At this time, if no subject candidate is detected before (on the left side of) the verb, the object selection processing unit 25 determines that the sentence is an imperative sentence, and excludes it from a subject verb matching error detection target.

  Finally, the correctness / incorrectness determination unit 16 performs a determination process on matching of the subject verb with respect to the verb and the subject candidate extracted for each segment (step S108), and the result is output.

  Note that the order of normalization processing of input sentences is not essential, and the processing order can be changed as long as it does not affect the subsequent processing.

  The present inventor has determined the accuracy of the subject verb matching error detection by the matching error detection device 12 for a large number of English sentences written by Japanese learners of English (university students) including those whose subject verbs are not matched correctly. When we conducted an experiment to evaluate, we found that the subject verb matching error can be detected more accurately than the conventional method using syntax analysis.

  As described above, in addition to the mode in which the coincidence error detection device 12 is provided in the server, it is possible to cause the computer to function as the coincidence error detection device 12 by installing the program in a computer owned by each user. It is.

  In addition, the present invention can be used not only for the English learning system 10 but also for purposes other than English learning, such as checking translations in English.

  In addition, the configuration of each part of the apparatus in the present invention is not limited to the illustrated configuration example, and various modifications are possible as long as substantially the same operation is achieved.

DESCRIPTION OF SYMBOLS 12 Match error detection apparatus 14 Input sentence analysis means 15 Subject verb candidate extraction means 16 Correct / incorrect judgment means 22 Input sentence normalization process part 23 Extraction process part 25 Target selection process part 26 Parallel noun analysis process part 27 Input sentence division | segmentation process part 28 Unnecessary Phrase deletion processing unit 29 Inverted syntax conversion processing unit

Claims (5)

A subject verb matching error detection device for detecting a matching error between a subject and a verb in an input sentence in English,
Perform part-of-speech analysis and phrase analysis of the input sentence, specify words, phrases and clauses constituting the input sentence, part-of-speech information related to the type and number of parts of speech of the word, and phrase and clause information related to the type of phrase and clause Using the result of the input sentence analyzing means, and extracting the verb in the verb phrase for the input sentence based on the rules stored in advance without performing syntax analysis, and the verb The subject verb candidate extracting means for extracting a noun in the noun phrase existing on the left side of the subject as a subject candidate, and the personality and number of each of the verb and the subject candidate extracted by the subject verb candidate extracting means are stored in advance. An apparatus for detecting coincidence errors of a subject verb, comprising: correctness / incorrectness determination means for determining whether or not the subject candidate and the verb match based on the rules.   In the correctness / incorrectness determination means, if the extracted verb does not match the subject name and number for all the subject candidates extracted corresponding to the verb, the matching error exists in the input sentence. The subject verb matching error detection apparatus according to claim 1, wherein the subject verb matching error detection apparatus is determined.   The subject verb candidate extraction means includes an input sentence normalization processing unit that performs normalization processing on the input sentence to improve extraction accuracy of the subject candidate and the verb, and normalization processing by the input sentence normalization processing unit The subject verb matching error detection apparatus according to claim 1, further comprising: an extraction processing unit that extracts the subject candidate and the verb from the input sentence after being processed. The input sentence normalization processing unit specifies the type of the input sentence and performs selection processing for the sentence to be detected, and the number of words constituting a noun phrase in which nouns are arranged in parallel A parallel noun analysis processing unit that performs processing for adjusting the input sentence, an input sentence division processing unit that performs processing for dividing the input sentence at a predetermined position based on the analysis result of the input sentence analysis unit, and the input sentence For an unnecessary phrase deletion processing unit that performs a process of deleting a phrase that cannot be a subject candidate, and an inversion syntax of a subject and a verb starting from Thee, the word order of the subject candidate and the verb is rearranged as in a normal syntax An inversion syntax conversion processing unit for processing,
The subject verb matching error detection apparatus according to claim 3, wherein the extraction processing unit extracts the verb and the subject candidate for each part divided by the input sentence division processing unit. A program for detecting a matching error of a subject verb for causing a computer to execute a process of detecting a matching error between a subject and a verb of an input sentence in English,
Perform part-of-speech analysis and phrase analysis of the input sentence, specify words, phrases and clauses constituting the input sentence, part-of-speech information related to the type and number of parts of speech of the word, and phrase and clause information related to the type of phrase and clause Using the result of the input sentence analyzing means, and extracting the verb in the verb phrase for the input sentence based on the rules stored in advance without performing syntax analysis, and the verb The subject verb candidate extracting means for extracting a noun in the noun phrase existing on the left side of the subject as a subject candidate, and the personality and number of each of the verb and the subject candidate extracted by the subject verb candidate extracting means are stored in advance. A program for detecting a matching error of a subject verb for causing the computer to function as correctness / incorrectness determination means for determining correctness / incorrectness regarding the match between the subject candidate and the verb based on the rule.

Images