JPH04558A - machine translation system - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention] Technical field The present invention relates to a machine translation system.

BACKGROUND ART Pronounization in English includes forward anaphora and back anaphora. An example of the former is “Tom 5aid that he
ilashungry," and the antecedent of the pronoun "he" is 210m in front. On the other hand, an example of the latter is "The woman who hires"
d him will see Mortontomo
rrow, ” and the pronoun I+ 1., i mI+
The antecedent of is "Morton" which comes after.

With conventional English-Japanese machine translation, when an English sentence with a forward anaphoric form is translated into Japanese, it does not look so unnatural as Japanese, but in the case of a backward anaphoric form, it is translated as ``I hired him.'' The lady will meet Morton tomorrow.''
It would be unnatural in Japanese to interpret that ``he'' and ``R Morton'' are the same person. This is because the phenomenon of backward anaphora within a single sentence is rare in Japanese.

Until now, regarding English pronouns, a method has been announced in which when a nominative personal pronoun and its possessive case exist in the same clause, the possessive case is not translated into Japanese.
This method has the disadvantage that it cannot handle the phenomenon of backward anaphora across nodes, as seen in the upper part.

In addition, in conventional machine translation systems, when a nominative personal pronoun and its possessive case exist in the same clause, the possessive case is not translated into Japanese, or backward anaphora is used to process pronouns. In such cases, a method has been proposed in which the pronoun and its antecedent, the complete noun phrase, are exchanged and translated into Japanese (Japanese Patent Application No. 1-22
No. 019). However, depending on the target language (e.g.
In some languages (such as French and Spanish), it may be necessary to move pronouns placed in a certain order before finite verbs, a constraint that has never been addressed before.

Purpose The present invention has been made to solve the above-mentioned drawbacks, and focuses on the difference between anaphoric phenomena in English and anaphoric phenomena in Japanese, and in particular eliminates unnatural Japanese translations of back anaphora. To provide a machine translation system that can perform translation, and also to check whether pronouns other than the nominative and possessive pronouns exist in the translated text at the time of translation, and if there are multiple pronouns, to use a predetermined surface layer stored in the translation dictionary. This was done with the aim of providing a machine translation system that can rearrange verbs in the correct order according to constraints and move them before finite verbs.

Structure In order to achieve the above object, the present invention checks whether backward anaphorization is performed in an English sentence, and when it is performed, replaces the pronoun and its antecedent, the complete noun phrase, and converts it into a Japanese sentence. To translate as a word translation, or
an input section, an original text storage section that stores the original text to be translated that is input from the input section, a translation dictionary section that stores knowledge information used for translation processing, and a translation dictionary section that uses the contents of the translation dictionary section to perform the input processing. a translation unit that executes translation processing of the original text, a translation storage unit that stores the translated text obtained as a result of the translation processing, and a display unit that displays the original text and the translated text; The translation unit includes an editing control unit that edits the displayed original text or translated text, and the translation unit rearranges the translated text in the correct order according to a predetermined surface constraint regarding pronouns other than the nominative and possessive, and rearranges the translated text before the finite verb. This system is characterized by having a pronoun preprocessing section that moves to.

Hereinafter, a description will be given based on an embodiment of the machine translation system of the present invention. FIG. 1(a) is a schematic configuration diagram of the English-Japanese machine translation system according to the present invention. This is an example of a syntax conversion method using semantic information. In other words, it analyzes the original English sentence, creates an English intermediate structure, converts it into an intermediate expression that can express Japanese phrases, and generates Japanese.In the figure, 1 is the morphological analysis section. , 2 is a word dictionary, 3
is an inflectional word form file, 4 is a first information file, 5 is a composition analysis section, 6 is a grammar rule, 7 is a first analysis rule, 8 is an English structure identification section, 9 is a second information file, 10 is the first 2
11 is a third information file, 12 is a conversion unit, and 13 is a fourth information file.

14 is a syntax generator, and 15 is a morpheme generator.

First, the morphological analysis unit 1 analyzes morphemes.

The basis of this is word-by-word analysis. That is, using spaces between words, special characters such as hyphens, commas, and periods, and differences in uppercase and lowercase letters.

In addition to identifying words, it also estimates sentences and special clauses. Word identification is performed by comparing the word form in the inflected word form file 3 created from the word dictionary 2 with the word in the input sentence. Furthermore, interpretation of numbers and numerical expressions, local analysis using proper nouns, etc. is also performed using the first information file 4 created from the word dictionary 2.

Next, the syntax analysis unit 5 performs syntax analysis. Here, in order to understand the structure of a sentence, we will use syntactic analysis to understand grammar rule 6.
This is done with reference to the first analysis rule 7 created from .

That is, all possible interpretations are extracted by sequentially applying grammar rules to find out which words form a phrase, which phrases modify which phrases or words, etc.

However, this processing alone cannot eliminate semantically incorrect analysis.

Next, from the results of such syntactic analysis, the English sentence structure identification unit 8 generates an English intermediate structure while using semantic features to eliminate semantically incorrect analysis. At this time, a second information file 9 created using the word dictionary 2 and a second analysis rule 10 created from the grammar rules 6 are used.

This structure uses a phrase structure parsing tree with the starting symbol S of the sentence as the top node.

After that, the conversion unit 12 converts the Japanese language into an intermediate structure using the third information file 11 and the grammar rules 6.

Then, using the fourth information file 13 and grammar rule 6,
Japanese words are generated by the syntax generation unit 14 and the morpheme generation unit 15.

Therefore, in the embodiment of the present invention, as part of the English structure identification unit 8, the presence or absence of back anaphora in an English sentence is checked, and if back anaphora is present, the pronoun and its antecedent, the complete noun phrase, are analyzed. , and sends the result to the converter 12. In this English sentence structure identification unit 8, in order to represent the structure of English, the S node is used as the top node in FIGS. 3(a) and 3(b).
A tree structure as shown in is obtained.

Therefore, the processing of backward anaphora according to the embodiment of the present invention will be explained with reference to the flowchart of FIG. 2(a). First, it is checked whether a pronoun exists in the sentence using the dictionary information of each word that makes up the sentence. If the pronoun does not exist, this process ends. If a pronoun is present, the gender, number, and noun that matches the pronoun (
clause) exists, and if not, the process ends. If a noun (phrase) that satisfies the above condition exists, this noun (phrase) is extracted together with the pronoun and stored in the storage unit along with the case information assigned to each noun (phrase).

After this, it is structurally examined whether the extracted pronouns and nouns (phrases) have a backward anaphora relationship. First, trace the nodes upward from the node that controls the pronoun in question, and when you encounter the first S node, check whether there is a node of the previously extracted noun (phrase) under this S node. .

In this case, linguistic knowledge has revealed that if there is a noun (phrase) node, there is no backward anaphora relationship, and if there is no noun (phrase) node, there is a backward anaphora relationship.

If the relationship of backward anaphora is established, the noun (phrase) in question
and swap pronouns.

At this time, the substitution is performed by converting each pronoun and its antecedent noun (phrase) stored in the storage unit to match the case information given to the noun (phrase).

In this way, when the phenomenon of back anaphora exists, processing is performed to replace one pronoun and its antecedent, and the conversion unit 1
2, the resulting Japanese translation will be more natural. In the English example shown in the previous example, it would be ``The lady who hired Morton will meet the waves tomorrow.''
This is the Japanese translation.

Now, “The fact that she 1ie
nt to Paris impressed 5usa
When translating the English sentence "n's boss," into Japanese, we will explain it with reference to the main structure shown in Figures 3(a) and (b).The structural analysis tree for this input sentence is shown in Figure 3(a). ).First, refer to the dictionary information and select the pronoun from the word string.
Find "she". This pronoun is third person singular and the gender is female.

Next, when searching for a noun (phrase) that has the same meaning as this pronoun in the previous word string, "5uSan'S" is found.

Here, the structural positional relationship between "she" and "5usan's" will be examined. If you follow the nodes that control "she"1 upwards, you will encounter a node named S2 in FIG. 3(a).

Therefore, it is investigated whether these 82 notes dominate "5usan's". In this case, 82 nodes are “5us
an’s” and “she” because it does not control
It turns out that SanlSI+ is in a backward anaphora relationship.

Next, according to the case information stored in the storage unit, "she" and II 5usanlsI+ are exchanged, since the antecedent "5uSan'S" is possessive, it is 71
Convert SheI+ to “her”, and also
Since e''' is in the nominative case, "5usan's" is "5u".
It is necessary to convert it into the form "san" and perform the substitution.

In this way, a structural tree like the 3rd iM (b) is completed. This structure will be sent to the next converter. This results in a natural Japanese translation such as ``The fact that Susan went to Paris surprised her boss.''

Next, the schematic configuration and processing of the English-Spanish machine translation system will be explained with reference to FIG. 1(b). This is an example of a syntax conversion method using semantic information.

In other words, the original English sentence is analyzed, an English intermediate structure is created, and this is converted into an intermediate expression that can express the Spanish phrase to generate the Spanish word. This is the same as the case shown in FIG. 3A, except that the morpheme generation unit 15 generates a Spanish word.

In the embodiment of the present invention shown in FIG. 1(b), as part of the processing in the syntax generation unit]4, it is determined whether or not there are pronouns other than the nominative and possessive in the translated text, and if there are multiple pronouns, the translation is translated. The verbs are rearranged into the correct order according to predetermined surface constraints stored in the dictionary, and then moved before the finite verb. In this syntax generation unit 14, the S node is set as the top' node to represent the structure of the Spanish language, as shown in FIG. 3(C).
.

The tree structure shown in (d) is changed.

Therefore, pronoun preposition processing according to an embodiment of the present invention will be explained with reference to the flowchart of FIG. 2(b). First, the dictionary information of each word that makes up the Spanish sentence is used to check whether a pronoun exists in the sentence, starting from the end of the sentence. If the pronoun exists, the process ends. If there are pronouns, pronouns other than the nominative or possessive are extracted based on their case information and stored in the storage unit in the order in which they occur from the end of the sentence, along with the personal information assigned to each pronoun.

Next, it is checked whether there are two or more pronouns stored in the storage section. If there are two or more, step 4
If there is only one, the process moves to step 6.

In step 4, the surface constraints regarding the arrangement of pronouns in Spanish that are described in the translation dictionary are retrieved and checked for consistency with the arrangement of pronouns previously stored in the storage unit. The surface constraints for Spanish pronouns are as shown in Table 1 below. Numbers in the diagram represent personal names.

Table 1 se: Extremely special pronouns unique to Spanish ■: First person to express second person ■23rd person Here, if the order of the personal pronouns stored in the storage unit meets the above constraints, process step 6. If not, adjust the word order to match the order of the personal names in Table 1 (
Step 5).

Next, it is checked whether the verb in the sentence is a finite verb. A finite verb is a verb or its form that is distinguished by person, number, tense, modal, aspect, etc. Therefore, imperative, present participle, and infinitive verbs cannot be considered finite verbs. First, it is checked whether the verb in question is in the imperative form using the dictionary information given to each verb in Spanish. If it is an imperative form, it is further checked whether it is an affirmative imperative form or a negative imperative form. (In Western languages, you can easily tell whether it is an affirmative command or a negative command from the word form. In languages where the word form is the same,
If the Neg node adjacent to the NP node at the beginning of the sentence is empty, it can be determined that it represents affirmation, and otherwise it can be determined that it represents negation. ) If the verb is in the imperative form, the process moves to step 7; if the verb is in the affirmative imperative form, the process moves to step 8; if it is in the negative form, the process moves to step 11. If it is not in imperative form, check whether it is in present participle form.

If it is in the present participle form, the process moves to step 8. If it is not in the present participle form, check whether it is in the indefinite form. If it is an amorphous shape, the process moves to step 8; otherwise, the process moves to step 11.

In step 8, the pronoun group obtained up to step 5 is added to the verb and sent to the morpheme generator.

On the other hand, in step 11, move the V note that controls the verb to the right of the NP node that controls the pronoun,
This result is sent to the morpheme generator so that a correct translated sentence can be obtained.

Imaichi “5usan explained it to
The case of translating the English sentence "me," into Spanish will be explained with reference to the main structure shown in Figures 3(c) and (d).The tree structure in the syntax generator of this input sentence is shown in Figure 3 ( c).

First, refer to the dictionary information and select the pronoun "lo" from the word string.
(Three people S) and "me" (-personal name) are extracted and stored in the storage unit as shown in Table 2 below.

There are two pronouns stored in the second table storage, so step 4
Then, the process of step 5 is executed, and these pronouns are rearranged into an order that conforms to the surface constraint "me '" l ol.

Here, it is checked whether the verb ``pxplico'' is a finite verb. As a result, it turns out that "eχρ1ico" is a finite verb. Therefore, the process of step 1] is executed to convert the structure tree into a structure tree as shown in FIG. 3(d), and the result is sent to the morpheme generator. In this way, in the end”
5usan me lo avpljco," is obtained. If the verb in question is not a finite verb, the process in step 8 will produce, for example,
”S＋ノsan pudō explicarnel
o.

We get the following Spanish sentence:

Efficacy As is clear from the above explanation, according to the present invention, as mentioned above, it is checked whether or not pronounization, which is unique to English, called back anaphora is performed in the original text, and if it is, it is detected. The pronoun and its antecedent are changed and sent to the conversion unit ↓, which reduces unnatural translations into Japanese due to differences in syntactic structure between English and Japanese sentences, and improves the quality of the translated text. Vine that contributes to.

Also, during translation, it is checked whether there are pronouns other than the nominative and possessive in the translated sentence, and if there are multiple pronouns, they are rearranged in the correct order according to predetermined surface constraints stored in the translation dictionary, and they are placed before the finite verb. Since the translation takes turns, it is possible to reduce the occurrence of unnatural translated sentences due to differences in syntactic structure between the input language sentence and the target language sentence, contributing to higher quality of the translated sentences.

[Brief explanation of drawings]

FIG. 1 is a schematic configuration diagram of a machine translation system according to the present invention, (a) is a schematic configuration diagram of an English-Japanese machine translation system, (b)
) is a schematic configuration diagram of the English-Spanish machine translation system, Figure 2 (a)
Figure 2(b) is a flowchart for processing back anaphora, Figure 2(b) is a flowchart for pronoun prefix processing, and Figures 3(a) and (b) are for processing English sentences. Diagram showing the structural analysis tree for Japanese translation, Figure 3 (C)
. (d) is a diagram showing a structural analysis tree when translating an English sentence into Spanish. 1... Morphological analysis unit, 2... Word dictionary, 3... Inflection word form file, 4... First information file, 5...
Syntax analysis unit, 6... Grammar rule, 7... First analysis rule, 8... English structure identification unit, 9... Second information file, 10... Second analysis rule, 11. - Third information file, 12... conversion section, 13... fourth information file, 14... syntax generation section, 15... morpheme generation section. Patent applicant Rico Co., Ltd. Figure 1 Figure (a) Figure 1 (b) Figure 1 (ol Figure (C)

Claims (1)

[Claims] 1. Check whether backward anaphorization is performed in the English sentence, and if so, replace the pronoun and its antecedent, the complete noun phrase, and translate into Japanese. A machine translation system characterized by: 2. An input section, an original text storage section that stores the original text to be translated inputted from this input section, a translation dictionary section that stores knowledge information used for translation processing, and a translation dictionary section that uses the contents of this translation dictionary section. a translation unit that executes translation processing of the input original text; a translation storage unit that stores the translated text obtained as a result of the translation processing; and a display unit that displays the original text and the translated text; and an editing control section that edits the original text or translated text displayed in the translation section, and the translation section checks whether pronouns other than nominative and possessive pronouns exist in the translated text, and if there are plural pronouns, the translation dictionary What is claimed is: 1. A machine translation system comprising: a pronoun preprocessing unit that rearranges pronouns stored therein into the correct order according to predetermined surface constraints and moves them before finite verbs.