JPH09325673A - Translation sentence correction supporting device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain the result which is similar to that which is manually produced by a prefessional. SOLUTION: A model answer is analyzed into plural templates of a word- phrase construction employing the knowledge of a professional. Then, the messages for the answer, which is most likely translated by a learner, are made in advance using the knowledge of the professional. These messages are read (S1 and S2) and then, the answer sentence is read (S3). Then, a check is made to determine whether the learner correctly uses the fundamental usage or not (S4). Then, a part of speech classification is made for each word of the answer sentence (S5) and lastly, the answer sentence is analyzed for its sentence structure (S6). After that, the sentence is evaluated for every template (S8) and the message having the contents corresponding to the evaluation is provided to the learner (S9 to S14). Since the answer sentence is evaluated for every template, the correction result is similar to the result manually produced by the professional.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a translation sentence correction support device for mechanically correcting a learner's answer sentence learned in a technical translator training course, and in particular, results of the same degree as those performed by experts. The present invention relates to a translation sentence correction support device capable of easily obtaining a translation.

[0002]

2. Description of the Related Art Generally, in technical translator training courses, unlike fields such as literature that emphasize so-called literary expressions, engineers and scientists do not allow ambiguity in their interpretation. The goal is to use a sentence that has the simplest syntax and that conveys the intention correctly to anyone, and the learning text given to the learner is usually structured so that such a sentence expression can be learned. Is.

That is, in the learning text, basic expression methods that all the students should at least learn are summarized as basic patterns, and examples for learning these basic patterns are given. , The students are
A method is adopted in which the answer text for this example is sent to the headquarters, and the answer text sent to the head office is corrected and sent back to the student, and the student checks the content of the correction and confirms his or her level.

[0004]

By the way, conventionally, in the correction of answer sentences, a method has been adopted in which a large number of answer sentences are distributed to a plurality of correctors and each corrector manually performs the correction. It takes a lot of time to correct,
It is not always easy to secure a corrector with a certain level of ability, and there is a problem that the corrector's contents vary depending on the corrector.

The present invention has been made in view of the above situation, and it is possible to easily obtain a result to the same extent as an expert does by human power, and there is no possibility that the correction content may vary. The purpose is to provide a device.

[0006] Another object of the present invention is to provide a translation sentence support device capable of ensuring immediate response and paperless.

[0007]

In order to achieve the above-mentioned object, the present invention provides a terminal device operated by a learner, and a correction device for correcting the answer text from the learner connected to this terminal device and presenting the result. In a translation sentence correction support device including a device, the correction device is a receiving means for receiving an answer sentence from a learner; an answer sentence storing means for storing the received answer sentence; and a basic usage is properly taken. A key pattern checking means for checking the answer sentence by a key pattern for checking whether or not the answer sentence is present; a part-of-speech searching means for performing a part-of-speech classification for all the words constituting the answer sentence; a syntactic analysis of the answer sentence by a syntactic analysis algorithm A syntactic analysis means for executing; storing an arbitrary type of template pattern formed by decomposing the model answer corresponding to the answer sentence into a plurality of templates of word or phrase structure The model answer storage means; compares each template pattern of the model answer with the syntactically analyzed answer sentence, detects which template pattern the answer sentence matches, and picks up a portion corresponding to each template of the answer sentence. A message storing means for storing a plurality of messages corresponding to each template of each template pattern in the model answer;
Correction means for comparing each template of the model answer and a part corresponding to each template of the answer sentence, and selecting a message corresponding to the comparison result; and a transmitting means for transmitting each selected message to the learner It is characterized in that it is configured from. Then, since the answer sentence and the model answer are decomposed into templates and compared for each template, the correction work can be easily performed at the same level as an expert manually. There is no risk of variations in the correction content.

The present invention also provides a terminal device and a correction device,
It is characterized in that the connection is made through a network communication means such as the Internet. As a result, promptness of response can be ensured and paperless can be achieved.

[0009]

DETAILED DESCRIPTION OF THE INVENTION The present invention will be described below with reference to the drawings. FIG. 1 shows a translation sentence correction support device according to an embodiment of the present invention.
Includes a terminal device 2 installed at a learner's home or the like and a correction device 3 installed at a head office or the like. These devices 2 and 3 are network communication means 4 such as the Internet.
Connected to each other via the
Is accessible.

As shown in FIG. 1, the correction device 3 includes a receiving means 5 for receiving the answer text of the learner sent from the terminal device 2, and an answer text storage means 6 for storing the received answer text. This answer text is checked by the key pattern checking means 7 with a key pattern for checking whether or not the basic usage is correctly taken, and the part-of-speech searching means 8 composes the answer text. For every word that you do, part-of-speech classification is done by the built-in dictionary.

As shown in FIG. 1, the answer sentences classified into parts of speech are parsed by the syntax analysis program in the syntax analysis means 9, and the answer sentences subjected to the syntax analysis are compared. In the means 10, it is compared with the model answer stored in advance in the model answer storing means 11,
As will be described later, it is detected which template pattern of the model answer matches, and a portion of the template pattern corresponding to each template is picked up.

That is, in the model answer, by referring to past correction examples, what is the core element of the translated text in the correction work is examined using the correction knowledge of an expert, and the unit is used as a template. It is decomposed into called sentence elements, and the whole is preset as one template pattern. As shown in FIG.
It is adapted to be stored in the model answer storage means 11.

If a plurality of types of translations may occur for one example, template patterns corresponding to all the translations are set,
Each of these template patterns is stored in the model answer storage means 11. Then, the comparing means 10 detects which template pattern of the model answer the answer sentence matches, and the part of the matched template pattern corresponding to each template is picked up from the answer sentence. .

The corrected answer sentence is corrected in the correction means 12 based on the information from the message storage means 13.

That is, the message storage means 13 corresponds to each template of each template pattern in the model answer, for example, a message for the best choice, a message for an acceptable word / phrase choice, and a message for an incorrect choice. Etc. are respectively stored, and when the content of the part corresponding to the template of the answer sentence is, for example, the best choice, the correction means 12 selects the message for the best choice. This message is selected for each part of the answer sentence corresponding to each template, and each selected message is sent to the terminal device 2 via the sending means 14 as shown in FIG. .

Next, the structure of the correction device 3 will be specifically described. First, the key pattern checking means 7 will be described. As mentioned above, learning texts are given in advance to learners of technical translator training courses, etc. In this learning text, for example, in the case of Japanese-English translation, an English sentence describing the knowledge of the specialized field is written. The usages and examples of are shown as basic usages, and this basic usage is the content that all learners should at least learn to use correctly. Therefore, the key pattern is determined based on this basic usage.

The symbols necessary for setting the key pattern are defined as follows, for example. Symbols used in key pattern setting ・ ∧ Indicates the beginning of a sentence.・ $ Indicates the end of a sentence. * * Represents repetition of 0 or more words. -# H Indicates that the part of speech is H. -[A | B | ... | N] Indicates one of A, B, ..., N. Other rules ・ A ∧ is required at the beginning of the key pattern.・ A $ is required at the end of the key pattern. -At least one word information or part-of-speech information is required in the key pattern.・ Repeat symbol ++ must not be duplicated.・ If there is more than one information in [] that matches the answer sentence, select the leftmost one.

An example is now defined as "a" scientific substance "is a substance obtained by a chemical method. , The key pattern of the answer sentence is set as follows in accordance with the procedure for setting the key pattern. ∧ + [[is | are] defined as | [wa
s | were] defined to be] + $ This setting is defined when the be verb is in the present tense.
If ed as is more appropriate and the be verb is past tense, it means that defined to be is desirable. Then, the answer sentence is checked by this key pattern.

Next, the part-of-speech searching means 8 will be described.
The answer sentence whose key pattern has been checked is classified by a built-in dictionary for all the words constituting the sentence. Specifically, it is first checked whether the word used in the answer sentence is in the dictionary. If it does not match the word in the dictionary, the ending of the word in the answer sentence and the word in the dictionary are examined, and the word in the dictionary changed to the plural form or the past tense matches the word in the answer sentence. Can be investigated. If the word does not match the word in the answer sentence even if the ending is changed, it is checked from the irregular change data whether there is a word in the answer sentence, and if there is, a part of speech is searched from the original form of the word.

Next, the syntax analysis means 9 will be described.
When the part-of-speech information of each word in the answer sentence is obtained by the part-of-speech search, ATNG (Augmented Trans) by top-down vertical search using, for example, ATN.
The syntax analysis of the answer sentence is performed by using the "Edition Network Grammar".

The ATN is used as a state transition of a finite automan. A finite automan is a structure that models a calculation as a transition from one with a finite number of states to another. In this ATN, various checks can be performed in the process of this state transition. The following grammatical rules are used to check the answer sentence in the syntactic analysis. S → NP VP NP → + det * adj n + NP + PP VP → be vt + PP PP → prep NP Here, “+” indicates that it can be omitted, and “*” indicates 0 or more repetitions.

Next, an analysis example will be described. Answer sentence is Chemicals are defined as
substancesobtained by che
medical methods. An example of analysis in the case of is as follows. Now, it is assumed that the word of the answer sentence indicates Chemicals at the start of the analysis. The grammar rule selects S → NP VP. The NP on the right side of this rule is NP → + det * adj
Replace with n + NP + PP. Also, chemica
It is known that the part of speech of ls is n (noun). Looking at the grammar rules, the beginning of the right side is + det. This is c
Since it does not match n of chemicals, this + det is omitted and the process moves to the next * adj. At this time, the word remains as it is without moving. Next, * adj and chemicals
Since it does not match even when compared with the part of speech n of, the * adj is repeated 0 times, and the process moves to the next n. C here for the first time
Matches n in chemicals. chemicals
Since there is a match with, it moves to the next word are.

The part of speech of this word is be representing a be verb. Subsequently, the grammar rule moves from n to the next + NP. NP
Is again replaced by + det * adj n + NP + PP. Therefore, the current grammar rule points to + det.
+ de because the part of speech be of be and + det do not match
When t is omitted, the process moves to the next * adj, but since this also does not match, it is assumed to be 0 or more times of repetition. Moving to the next step, the grammar rule is n and the word part of the answer sentence is part of speech, so they do not match. NP → + det * adj n + NP + P
Since the grammatical rules of P do not match, a grammatical rule having another NP on the left side is searched, but since this also does not exist, + NP is omitted. Then move to the next + PP.

PP can be replaced by PP → prepNP, but since the parts of speech of prep and are do not match, a grammatical rule having PP on the left side is searched for, but since there is no other case, + PP is also omitted. This will match the initially selected grammar rule. Up to here, NP →
By + det * adj n + NP + PP, Chemi
This means that cals can be analyzed. Then, it is only necessary to move to the analysis of VP and the like in the same procedure below, and when the grammar rules can be traced to the end of the sentence from the top down, the grammar has no grammatical error.

Next, the comparison means 10 will be described. The answer sentence syntactically analyzed by the syntactic analysis unit 9 is compared with the model answer stored in the model answer storage unit 11 by the comparison unit 10.

Here, the template pattern of the model answer stored in the model answer storage unit 11 is of two types shown in Tables 1 and 2, while the answer sentence is "A chemicals substance" we.
re defined to be a thing g
ot by chemical method. Then, the processing in the comparison means 10 is as follows.

[0027]

[Table 1] ---- means that any combination of T1 and T2 may be selected.

[0028]

[Table 2]

The comparison means 10 first detects which type of template pattern of the model answer the answer sentence is. From this, it can be seen that it matches the template pattern of the type shown in Table 2.

Next, the comparison means 10 picks up a portion corresponding to each template T1 to T5 of the template pattern of Table 2 from the answer sentence. As a result, the answer sentence is decomposed into templates similar to the template pattern in Table 2.

Next, the correction means 12 will be described. As described above, the message storage unit 13 stores a message or the like for the best selection in association with each template of each template pattern in the model answer, and the correction unit 12 includes a portion corresponding to each template of the answer sentence. Determine what the content of is. Then, when the content is, for example, the best selection, the correction means 12 reads the message for the best selection from the message storage means 13 and outputs this as the correction result.

An example of the correction result is shown below. ““ A chemical substance ”” □ The chemical substance is a chemical or chemical
Use cals. □ There is a problem in the position of "or". Do not put an article between "and" or between "and". "Were defined to be" □ Single and plural do not match. Thing ”□ The word that seems to correspond to“ substance ”is not properly selected. “Got by” □ The word that seems to be “obtained by” is not properly selected. “Chemical method” □ There is no article.

In the message storage means 13,
A more detailed message than the content of the message is stored in a tree format so that the learner can know the content as needed.

An example of more detailed message contents regarding the "substance" in the correction result will be shown below. "Material" BEST: substance ALLLOWABLE: material NOT ALLLOWABLE: product, thi
ng □ Select an appropriate word ・ Why "substance" is "material"
Is it more suitable? “Material” is a general word that describes all created elements, etc., whereas “substance” when applied in this case means a “substance” that is a particular chemical structure. .・ Why is "product" wrong? Chemical products are not treated as final products.
They are used in other industries, experiments, etc. For this reason, chemical products are intermediate rather than commercial products.・ Why is “thing” wrong? "Chemical" gives the image of an artificially made pure substance for use in experiments, industry. However, "thing" means a wide range. “Thing” gives an image of a concrete item such as a bottle or an umbrella. □ Depending on the singular or plural of the subject, the singular or plural. □ Use of Articles When used in singular form, the indefinite article “a” is required.
“The” is not appropriate in both singular and plural forms.

FIG. 2 is a flow chart showing a correction method by the translation sentence correction support device 1. The operation of the present invention will be described below with reference to FIG.

As shown in FIG. 2, first, in step S1, each template pattern of the model answer is read, and in step S2, a message corresponding to each template and a more detailed message are read. After that, in step S3, the answer sentence from the learner is read.

Then, in step S4, the answer sentence is checked by a key pattern for checking whether or not the basic usage is properly taken.

Next, in step S5, part-of-speech classification is performed for all the words constituting the answer sentence while referring to the built-in dictionary.

Next, in step S6, the answer sentence after the part-of-speech classification is parsed by a syntactic analysis program,
Next, in step S7, the answer sentence and the model answer are compared. Then, it is detected which template pattern of the model answer the answer text corresponds to, and which part of the answer text corresponds to each template of the template pattern.

Then, in step S8, the contents of the part of the answer sentence corresponding to each template are evaluated, and in step S9, the message corresponding to the evaluation is read from the message storage means 13. Then, in step S10, the read message content is sent to the learner as a correction result and displayed on the CRT of the terminal device 2.

Next, in step S11, it is determined whether or not the learner needs more detailed message contents, and if so, the more detailed message is sent to the message storage means 13 in step S12.
At the same time, the detailed message thus read is displayed on the CRT of the terminal device 2 in step S13. Then, in step S14, it is determined whether the learner needs another detailed message,
If necessary, the steps S12, S13
Is repeated.

However, since the model answer is decomposed into a plurality of templates in advance using the knowledge of the expert and the answer sentence is evaluated for each template to obtain the correction result, the expert It is possible to easily obtain the same correction results as those performed manually. In addition, the message corresponding to the evaluation of the answer sentence is previously stored in the message storage unit 13.
Since it is only necessary to select the one stored in the table, the correction result can be promptly shown to the learner, and there is no fear that the correction content varies.

In the above embodiment, the case where the terminal device 2 and the correction device 3 are connected via the network communication means 4 such as the Internet has been described. However, for example, a learner attends a technical translator training school. In such a case, it is possible to directly connect the two devices 2 and 3 in the training center.

Further, in the above-mentioned embodiment, the correction of the Japanese-English translation has been taken as an example, but the same applies to the case of the English-Japanese translation, and the same applies to the translation between other languages. And the same effect can be expected.

[0045]

As described above, according to the present invention, in a translation sentence correction support device having a correction device including a terminal device and a correction device, a receiving means, an answer sentence storing means, a key pattern checking means, A part-of-speech search means, a syntax analysis means, a model answer storage means, a comparison means, a message storage means, a correction means, and a transmission means are provided, and the answer sentence and the model answer are decomposed into templates for each template. Since they are compared and evaluated, the correction work can be performed at the same level as an expert does by human power, and there is no fear that the correction contents will vary.

The present invention also includes a terminal device and a correction device,
Since the connection is made via a network communication means such as the Internet, the response can be secured immediately and can be made paperless, and the learner can use the time zone convenient for the correction result. Can be obtained.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram showing a translation sentence correction support device according to an embodiment of the present invention.

FIG. 2 is a flow chart showing a correction method by the apparatus of FIG.

[Explanation of Codes] 1 translation sentence correction support device 2 terminal device 3 correction device 4 network communication means 5 receiving means 6 answer sentence storage means 7 key pattern checking means 8 part-of-speech searching means 9 syntax analyzing means 10 comparison means 11 model answer storage means 12 correction means 13 message storage means 14 transmission means

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] April 8, 1997

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] 0018

[Correction method] Change

[Correction contents]

The example is now defined as "a chemical substance" is a substance obtained by a chemical method. , The key pattern of the answer sentence is set as follows in accordance with the procedure for setting the key pattern. Λ + [[islare] defined asl [wa
slwere] defined to be] + $ This setting is defined when the be verb is in the present tense.
If ed as is more appropriate and the be verb is past tense, it means that defined to be is desirable. Then, the answer sentence is checked by this key pattern.

[Procedure amendment 2]

[Document name to be amended] Statement

[Name of item to be corrected] 0024

[Correction method] Change

[Correction contents]

PP can be replaced by PP → prepNP, but since the parts of speech of prep and are do not match, a grammatical rule having PP on the left side is searched for, but since there is no other case, + PP is also omitted. This will match the initially selected grammar rule. Up to here, NP →
By + det * adj n + NP + PP, Chemi
This means that cals can be analyzed. Then, it is sufficient to move to the analysis of VP and the like in the same procedure below, and when the grammar rules can be traced from the top down to the end of the sentence
The sentence has no grammatical error.

Claims (2)

[Claims] 1. A translation sentence correction support device comprising: a terminal device operated by a learner; and a correction device which is connected to the terminal device and corrects an answer sentence from the learner and presents the result, The correction device includes a receiving means for receiving an answer sentence from a learner; an answer sentence storing means for storing the received answer sentence; an answer by a key pattern for checking whether or not the basic usage is properly taken. A key pattern checking means for checking a sentence; a part-of-speech searching means for performing a part-of-speech classification for all words constituting the answer sentence; a syntactic analysis means for performing a syntactic analysis of the answer sentence by a syntactic analysis algorithm; an example corresponding to the answer sentence Model answer storage means for storing an arbitrary type of template pattern obtained by decomposing the answer into a plurality of templates of word or phrase structure; template templates for each model answer Comparing means for comparing the turn and the parsed answer sentence to detect which template pattern the answer sentence matches, and for picking up the part corresponding to each template of the answer sentence; each template pattern in the model answer And a message storing means for storing a plurality of messages corresponding to each template; and a correction for selecting a message corresponding to the comparison result by comparing each template of the model answer and the part corresponding to each template of the answer sentence. A translation sentence correction support device comprising: a means; and a transmission means for transmitting each selected message to a learner. 2. The translation sentence correction support device according to claim 1, wherein the terminal device and the correction device are connected via a network communication means such as the Internet.