JP3388393B2 - Translation device for tense, aspect or modality using database - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a translation device using a database. In particular, the tense, aspect, or modality of sentence endings that becomes a problem when translating from a language in which tense is expressed at the end of the sentence, such as Japanese, to another language such as English. Used when translating.

[0002]

2. Description of the Related Art A conventional example of a conventional translation method in a translation device using a database is the first conventional technique, and a conventional example of a translation method in a translation device using a database based on an example. The second prior art will be described below, and the third prior art will be described below as a third prior art in which the number of matching character strings counted from the end of a sentence is used as a method of checking the similarity between examples.

First, a first conventional technique of a translation method of a translation device using a database is shown in the flowchart of FIG. The flowchart of FIG. 2 shows the following four steps.

Conventional Japanese-to-English translation of sentence end expressions has been done by rules created manually. Therefore, first of all, it was necessary to perform the following work.

1) Prior to analysis, a rulebook is manually created in advance. For example, if the continuous form + verb "Iru" is used, the aspect is "progressive phase". Such rules are created for other combinations to create a rule set. Also, for tense, aspect, or modality, a rulebook will be created. When a rulebook is created by such a person, deficiencies in the rules remain, and it is necessary to constantly maintain and refine the rules.

Next, as an analysis work, 2) morphological analysis and syntactic analysis of the input sentence for translation of the procedure 1 input sentence in the analysis are performed. For example, when "I'm hoping." Is the input sentence, the following result is obtained.
Hope <noun> is <particle> say <verb><synonym><verb>

[0007] Here, if the morphological analysis unit or the syntactic analysis unit is changed, the above rule set is also affected, and in order to maintain an appropriate translation, there is a point to be changed in the above rule set. Will end up.

3) Procedure 2 in analysis The rule is collated with the result of morphological analysis or syntactic analysis to determine the tense (tense), aspect (phase) or modality (modality).
In the above case, since the sentence end expression is in the form of <conjunctive form> + verb “Iru”, the “progressive phase” is determined by the rule created in advance.

Then, to compose the whole sentence as follows,
Perform synthesis work. 4) Procedure 3 in the analysis The parts other than the translation of the tense, the aspect or the modality may use any of the conventional well-known translation methods, and the translation may be performed by using any one of them. The modality is translated by the above method, and by synthesizing them, the translation of the entire sentence is completed.

The method of translating according to the manually created rule shown here has a drawback in that a large amount of human resources must be invested in the maintenance of the rule.

Next, a second prior art as a prior example of a translation method in a translation device using a database based on an example will be shown.

The method of the present invention also utilizes a database of examples and is classified as an example-based method. As a conventional example in which the method based on this example is used for Japanese-English translation, a report (Eiichiro Sumita, Hi
toshi Iida, and Hideo Kohya
ma, Translating examples: A
new approach tomachine tra
nslation, The third Interna
regional Conference on Theore
mechanical and Methodological Is
suesin Machine Translation
of Natural Language, No.3,
(TMI, 1990), pp.203-212). However, the problem with this technology is "A's B",
In the Japanese-English translation of "A no B", the semantic information of the noun A and the noun B was used in a complicated combination. This study differs from the present invention in the following points.

1) Configuration of a database used corresponding to the first database of claim 1, 2) Method of evaluating similarity between examples, 3) Position of interest in one example.

Finally, as a third conventional technique, a conventional example in which the number of matching character strings counted from the end of a sentence is used as a method of checking the similarity between examples will be shown.

Also in the present invention, the number of matching character strings counted from the end of a sentence is used as a value (similarity) for evaluating similarity, but this method itself has a conventional example. Study on complementing abbreviations at the end of sentences (Maki Murata, Makoto Nagao, Complementing verb abbreviations using surface expressions and examples in Japanese sentences, Journal of Linguistic Processing, Vol. 5, No. 1, (1998)) There is an example base usage that uses the number of characters in the end-of-sentence matching character string as the degree of similarity.

The conventional example and the present invention are different in the following points. 1) In the conventional example, the target problem is complementation of abbreviations, not translation between different languages as in the present invention. 2) The structure of the database used corresponding to the first database of claim 1 is different in language and item.

The above-mentioned prior art is only partially in agreement with the method of the present invention, and it is not easy to think of the translation method of the present invention by simply combining these prior arts. it is obvious.

[0018]

In the translation method in the translation device using the conventional database, the conventional translation of the sentence end expression has been done by the rule created manually. However, the method of translating by a rule created by hand has a drawback in that it is necessary to invest a large amount of human resources for the maintenance of the rule to improve the translation accuracy.

The present invention has been proposed in view of the above, and uses a database that does not require manual creation of a rule set and can improve translation accuracy without knowledge of translation using the database. An object is to provide a translation device .

[0020]

The means used in the present invention to achieve the above object will be described with reference to the flow chart of FIG. The method according to the present invention will be briefly described below.

1) Prior to analysis, a database is prepared in advance, which collects examples of the first language and corresponding examples of the second language. At this time, the classification of the tense, aspect or modality of the second language example is added. This assignment may be performed manually, or an already well-known morpheme and parsing system can be used as an aid.

2) Procedure 1 in analysis In the search for translation of the input sentence, an example in which the matching character string from the end of the sentence is the longest,
Search from the above database. The well-known binary search method can be used as the search method.

3) Procedure 2 in the analysis, the sentence of the verb part on the English translation side of the example extracted from Procedure 1 in the confirmation of the translation,
The classification of aspect or modality is established as the tense, aspect or modality of the input sentence.

4) Procedure 3 in analysis In the method of constructing a translated sentence, any of conventional translation methods may be used for the portion other than the translation of the tense, aspect or modality,
Translate using any of them, and translate tense, aspect or modality by the above method,
The translation of the entire sentence is completed by synthesizing these.

Therefore, in order to achieve the above object, the invention according to claim 1 is a translation device for a tense, an aspect, or a modality using a database from a first language to a second language. An example consisting of a plurality of examples belonging to a language and a plurality of examples belonging to a second language and belonging to each first language has at least one correspondence with an example belonging to the second language. , Tense,
It is equipped with a first database characterized by having aspect or modality information added, and it has a first example belonging to a first language and a first language of the first database.
Using the number of common character strings that are continuous from the end of the sentence with the second example to which it belongs, derive a value that evaluates the similarity with the second example that belongs to the first language of the first database. comprising means, said means were evaluated (1) the similarity value
, The higher the similarity, the higher the similarity ,
Select a second example group belonging to the first language for the first example belonging to the first language by a first method of selecting a predetermined number from the first database, (2)
From the correspondence of the examples belonging to the first language to the examples belonging to the second language, the first example group belonging to the second language is selected from the first database, and (3) the selected second The tense, aspect or modality representing the first example group belonging to the language of is determined by the second method of determining the tense, aspect or modality by the majority of the tense, aspect or modality of each example, ( 4) the determined Tense, aspect or modality, the second example of the translation of Tense belonging to the first language, is used as an aspect or modality, and characterized by comprising a structure that,
We propose a way to properly translate tense, aspect or modality in an example-based manner.

The invention according to claim 2 is a database.
-Based tense, aspect or modality
Is a translation device related to the
The feature is that a morpheme is analyzed to recognize a morpheme, a thesaurus classification number is attached to each morpheme, and the first example belonging to the first language converted into a configuration according to the thesaurus classification number is similar to the first example. It is characterized in that the number of common character strings continuous from the end of the sentence is used between the converted second database and the second example belonging to the first language .
The similarity between the second example in the first language and
It is equipped with a means for deriving a valuated value.

[0027]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described in detail below. First, the first embodiment will be described using Table 1.

[0028] now, "he my acquaintance's" think that to translate the tense of. At this time, it is assumed that Table 1 shows a collection of a large number of Japanese-English translation pairs, "He is my acquaintance", and 10 examples from the top that have many string matches from the end of the sentence. .

[0029]

[Table 1]

The similarity of the table shows the number of matches a string from the end of the sentence. Also, the k-nearest neighbor method is used here. The k-nearest neighbor method is a method of obtaining a majority rule of k examples taken in order from the highest similarity, instead of using one most similar example.

[0031] it is necessary to perform majority with all similarity is equal example, regardless of the value of k if there are examples similarity is equal. Furthermore, because of the simplicity of processing here,
Only 10 examples will be examined at most.

Further , in Table 1 above, the classification column is obtained from the corresponding verb phrase in the English sentence, but this portion may be automatically performed using a well-known processing program, or it may be performed in the database. When creating, the classification may be manually entered in advance.

[0033] First, consider the case of k = 1. At this time, one example with the highest similarity is used for analysis, but since the first and second examples have the same similarity, analysis is performed using the first and second examples. When a majority decision is made with this, it is said that the classification is 1 for "currently completed" and 1 for "current", and when the opinion is found, it is decided that the classification that went up earlier is the solution, and the solution goes up first. It becomes "currently completed" and becomes an incorrect answer.

[0034] Next, consider the case of k = 3. At this time, the three with the highest degree of similarity are selected, but since the third example and thereafter have the same degree of similarity, all ten examples are used. If a majority decision is made with this, the opinion is divided as "currently completed" is 2 and "currently" is 8, but there is a large number of "currently", which corresponds to the correct "currently" and is correct.

[0035] Next, as well as 10 in the case of k = 5,7,9
All the examples are used and the solution is "present", which is also correct.

In this problem, the system is such that when k = 1,
This means that an incorrect solution is output, and a correct solution is output when k = 3, 5, 7, 9. Regarding the value of k, an appropriate value may be selected when actually manufacturing the device. The value of k obtained by this method is usually an odd number for the sake of majority, and 3 or 5 is often sufficient. As the number of database applications increases, smaller k values can be used.

Next the second embodiment will be explained with reference to Table 2. Now consider translating the tense "He's my acquaintance". At this time, for a database that collected a large number of Japanese-English translation pairs, "He is my acquaintance" was collected from the top 10 examples of many string matches from the end of the sentence. Suppose

[0038]

[Table 2]

[0039] Table 2 of similarity recognizes the morphemes by performing morphological analysis of the input sentence, for each morpheme assigned the classification number of the thesaurus, and provide an input sentence is converted to the configuration according to the classification number of the thesaurus Also, prepare similar database for a large collection of Japanese-English translation pairs, and for these converted sentences, determine the number of common character strings that are continuous from the end of the sentence. Shows.

For the analysis, the k-nearest neighbor method will be used as in the first embodiment.

[0041] First, consider the case of k = 1. At this time, the analysis is performed using only the first example having the highest degree of similarity. In the first example, the classification is “currently completed”, which is an incorrect answer, unlike the correct answer “current”.

Next, consider the case of k = 3. At this time, the three with the highest similarity are selected, but the third example and 4
Since the number examples have the same degree of similarity, four examples up to the number example 4 will be used. When a majority decision is made with this, it is understood that the classification is 2 for "currently completed" and 2 for "currently",
The solution is now "completed", which is also an incorrect solution.

Next, consider the case of k = 5. At this time, the five with the highest degree of similarity are selected. However, since the fifth and subsequent examples have the same degree of similarity, all ten examples are used. If a majority decision is made with this, it can be understood that the classification is “currently completed” is 2, and “current” is 8, but the number is the “current”, which is a large number, which coincides with the correct “current” and is correct.

Next, k = 7,9 is the solution as well as all 10 of the example is used in the case of "current", and this is also correct
Becomes

In this problem, the system will output an incorrect solution when k = 1,3 and a correct solution when k = 5,7,9. Regarding the value of k, an appropriate value may be selected when actually manufacturing the device. For the value of k by this method, an odd number is usually desirable for the sake of majority, and 7 or 9 is often sufficient. Also in this case, a smaller value of k can be used as the number of database applications increases.

As shown in the above embodiment, in the method of the present invention, it is possible to improve the translation accuracy by maintaining the database in which the examples are collected. Therefore, it is possible to manually prepare the rule set. Maintenance is easy because there is no need to do it, and translation accuracy can be improved without knowledge of translation using a database.

[0047]

Since the present invention has the above-mentioned structure,
The effects described below can be achieved.

[0048] In the invention described in claim 1, enables translations based on examples, it is easy maintenance it is not necessary to the creation of Regulations manual, also endnote viewed
The number of consecutive common strings
As a result, the similarity can be easily evaluated, and the translation accuracy can be improved without any knowledge of translation using the database.

[0049] Further, in the invention according to claim 2, it will be able to evaluate the similarity using the similarity of the meanings, can now also appropriate translation from the mean.

[Brief description of drawings]

FIG. 1 is a flowchart showing a means for solving the problems in the present invention.

FIG. 2 is a flowchart showing a means in the prior art.

─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Hitoshi Isahara 558-2 Iwaoka, Iwaoka-cho, Nishi-ku, Kobe, Hyogo Prefectural Telecommunications Research Institute Kansai Branch (56) Reference JP-A-6-309352 (JP, A) (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06F 17/21-17/28 JISST file (JOIS)

Claims (2)

(57) [Claims] 1. A translation device using a database from a first language to a second language, comprising a plurality of examples belonging to a first language and a plurality of examples belonging to a second language, and each individual first The example that belongs to the language of has at least one or more correspondence with the example that belongs to the second language.
A first database having a first database characterized by adding aspect or modality information, and a first database belonging to a first language and a first database
Similarity between the second example belonging to the first language of the first database and the second example belonging to the first language by using the number of common character strings continuous from the end of the sentence comprising means for guiding the value obtained by evaluating, said means, the higher the value obtained by evaluating (1) the similarity
In order of high similarity, the second example group belonging to the first language with respect to the first example belonging to the first language is arranged in a descending order from the first database by a predetermined number <br / > Select by the first method of selecting, and (2) first example group belonging to the second language from the correspondence of the example belonging to the first language to the example belonging to the second language. (3) For the tense, aspect or modality representing the first example group belonging to the selected second language , the tense, aspect or modality is set for each individual example tense, aspect or modality. in determined by the second method of determining by majority, (4) the determined Tense, aspect or modality, the second example of the translation of Tense belonging to the first language, a Used as and aspect or modality
That, characterized in that it comprises a structure that, Tense using the database, the translation device relates aspect or modality. 2. A translation device using a database from a first language to a second language, comprising a plurality of examples belonging to the first language and a plurality of examples belonging to the second language, and each individual first The example that belongs to the language of has at least one or more correspondence with the example that belongs to the second language.
It is equipped with a first database characterized by the addition of aspect or modality information, recognizes morphemes by performing a morpheme analysis with a first example belonging to the first language, and identifies the thesaurus classification numbers for each morpheme. And a second example belonging to the first language in the first database, which has been converted into a structure based on the thesaurus classification number, and a second example similarly belonging to the first database. between, comprising means for guiding the value obtained by evaluating the similarity between the second example belonging to the first language, which comprises using a number of common string consecutive when viewed from the end of the sentence, said means (1) The higher the evaluation value of the similarity,
In order of high similarity, the second example group belonging to the first language with respect to the first example belonging to the first language is arranged in a descending order from the first database by a predetermined number <br / > Select by the first method of selecting, and (2) from the correspondence of the example belonging to the first language to the example belonging to the second language, the first example group belonging to the second language is first (3) For the tense, aspect or modality representing the first example group belonging to the selected second language , the tense, aspect or modality is set for each individual example tense, aspect or modality. in determined by the second method of determining by majority, (4) the determined Tense, aspect or modality, the second example of the translation of Tense belonging to the first language, a Used as and aspect or modality
A database-based translation device relating to a tense, an aspect, or a modality characterized by having a configuration .