JPH04263364A - Automatic translation method for natural language - Google Patents

Abstract

PURPOSE:To attain the translation method that can secure the translated word suitable to an object language even to such a predicate that has the different types of translation between a case where the predicate of an expression pattern of an input sentence is used for an ordinary sentence and a case where the preceding predicate is used as an adjective clause. CONSTITUTION:When a 1st natural language is translated into a 2nd natural language, a case where the predicate of an expression pattern of an input sentence is used for an ordinary sentence is registered in a coupling value pattern dictionary 7 together with a case where the predicate is used for an adjective clause. A language analyzing part 2 obtains the analyzing result of a sentence inputted from an input part 1 and then selects the expression patterns according to an adjective clause or a predicate decided from the analyzing result. Then a conversion generating part 4 composes a sentence and an output part 5 outputs the 2nd natural language.

Description

[Detailed description of the invention]

0001

[Industrial Application Field] The present invention relates to a natural language automatic translation method, and in particular, to automatically translate a first natural language sentence inputted by an input means into a second natural language sentence and output it. Concerning natural language translation methods.

0002

2. Description of the Related Art A conventional case of translating from Japanese to English will be explained. Previously, a dictionary was prepared in which the predicate ``Genkida'' was matched with the Japanese pattern ``《人》ga genki dai'' and the English pattern ``《人》be fine.'' A common method is to analyze words and convert from Japanese to English (Japanese Patent Application No. 232826/1982).

[0003] Here, the Japanese pattern "person" is "child"
, "Student", etc., represent a group of words representing humans. In the input sentence "The child is doing well.", "
Since ``child'' belongs to ``person'', it is judged to fit the above Japanese pattern. Translate “child” to “a chi”
ld is fine”.

[0004] Also, even when used as an adjective clause, such as "a cheerful child," the above Japanese pattern is used to determine whether or not the expression "the child is energetic" is valid. If it matches, use the English pattern according to the previous example (Patent Application No. 45271/1982). And once I said “a child who is fine”
After generating the internal expression , by degenerating the relative pronoun, we can change ``a fine chi'' to ``a fine chi''.
There is a known method of translating it as "ld".

[0005] Furthermore, by registering ``energetic'' → ``fine'' as an adjective in a bilingual dictionary, ``energetic child'' can be changed to ``a fine c'' without degenerating relative pronouns.
There is also a known method of translating it as "hild".

[0006]

[Problem to be Solved by the Invention] However, according to the bond valence pattern according to the conventional method, "A child is fine" when "the child is fine" and "healthy child" when "the child is healthy". Therefore, it is difficult to translate the predicate ``Genkida''. In other words, if the English pattern of the valence pattern is ``《人》be healthy,'' in contrast to the Japanese pattern of the valence pattern, ``《人》とは《人》is healthy,'' then it becomes ``The child is healthy.''
It also becomes “A child is healthy”.

Furthermore, according to bilingual dictionaries, there is no problem when an adjective clause consists of only a predicate, but it becomes difficult to deal with when it has other case elements. For example, in the case of ``a child whose parents are doing well,'' although ``cheerful'' is a predicate of an adjective clause, it needs to be treated as a normal predicate such as ``his parents are doing well.'' In other words, if a bilingual dictionary determines a fixed translation method just because it is a predicate of an adjective clause, it may result in an incorrect translation.

[0008]Furthermore, “impressive photos” are
In other words, if a Japanese adjective clause can be expressed with one English word, if you try to translate it as an adjective clause, you will need to check the English translation at the Japanese analysis stage. The problem is that it is inefficient.

[0009] Also, this problem concerns, for example, whether the word that appears in the position of ``...ga'' is different when ``Genkida'' is used as a predicate in a normal sentence and when it is used as a predicate in an adjective clause. , because they belong to the same group called ``people.'' However, since there are often differences in meaning and nuance between ordinary predicates and predicates in adjective clauses, it becomes a big problem to make precise translations. (A small dictionary for improving Japanese: Adjectives and Adverbs) provides examples of things that cause differences in nuance.
(written by Yoshiyuki Morita, published by Kodansha Gendai Shinsho). ) The present invention was created in view of the above points, and aims to provide a translation suitable for the target language even for predicates that are translated differently depending on whether they are used as a normal sentence or as an adjective clause. The purpose of this study is to provide an efficient natural language automatic translation method.

0010

[Means for solving the problem] A plurality of expression patterns of a second natural language and their selection conditions are defined for the keys of expression patterns of case elements and predicates in a first natural language, using predicate words as keys. The described bond valence pattern dictionary, an input means for reading an input sentence which is a first natural language sentence, a language analysis means for performing word segmentation and syntax analysis on the input sentence read by the input means, Based on the valence pattern dictionary, the predicate of the input sentence is selected from among multiple patterns of the second natural language that has multiple expression patterns, and if there is one that matches the selection conditions, the pattern conversion result is After translating the words of the first natural language into the second natural language based on the pattern conversion result,
Using a natural language natural translation device equipped with an output means for outputting a sentence in a second natural language, the predicate of the expression pattern of the first natural language is used in the normal way and as the predicate of an adjective clause. For the two types of cases in which the predicate is used, the expression patterns of the second natural language are registered in the valence pattern dictionary, and the pattern conversion means determines whether the predicate in the input sentence is used as a normal predicate. , or the predicate of an adjective clause, are added to the selection conditions of the selection function, and an expression pattern is selected.

[0011]

[Operation] Even if there is a difference in meaning or nuance between a normal predicate in the first natural language and a predicate in an adjective clause, the present invention allows the predicate in the expression pattern of the first natural language to be used normally. Expression patterns can be registered in the valence pattern dictionary for two types: when used as an adjective clause and when used as a predicate in an adjective clause, so the selection function of the pattern conversion means can be used to translate into a form suitable for a second natural language. It becomes possible. At that time, there is no need to be aware of the translation format into English at the stage of Japanese analysis, so the language analysis means on the analysis side and the language conversion generation means on the generation side each perform processing according to the language. It can be carried out. Further, pattern registration for different translations may be performed by registering only the second natural language pattern for use as an adjective clause.

0012

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows a basic block diagram of an embodiment of the present invention. The figure shows a Japanese to English translation system as an example. Input unit 1 reads Japanese sentences. The language analysis unit 2 performs word decomposition and syntax analysis of the read Japanese sentence. The pattern conversion unit 3 refers to the valence pattern dictionary 7 based on the linguistic analysis result and converts the input Japanese predicate into English. At this time, if any of the plurality of English expression patterns matches the selection conditions, that expression pattern is selected. The conversion generation unit 4 prepares English sentences as natural sentences. The output unit 5 outputs a well-organized sentence. The analysis dictionary 6 is used by the language analysis section 2. The bond valence pattern dictionary 7 describes expression patterns of Japanese case elements and predicates, and corresponding English expression patterns and their selection conditions using predicate words as keys. It is assumed that the conversion generation dictionary 8 includes a bilingual dictionary of words.

First, in the natural language translation apparatus, a Japanese sentence inputted from an input section 1 is word-analyzed and syntax-analyzed by a language analysis section 2 using an analysis dictionary 6 . Next, the pattern converter 3 uses the valence pattern dictionary 7 to determine the translation form of the Japanese syntactic pattern into English for each predicate. The conversion generation section 4 uses the conversion generation dictionary 8 to convert words from Japanese to English and generate English sentences for words other than the predicates processed by the pattern conversion section 3, and outputs them from the output section 5.

FIG. 2 is a diagram showing the structure of a bond valence pattern according to an embodiment of the present invention. The bond valence pattern dictionary 7 will be explained together with the operation of the pattern converter 3.

First, if the input Japanese sentence is "The child is energetic," the valence pattern dictionary 7 is searched using the predicate "Genki da" as a key to obtain the Japanese pattern "The person is energetic." . Since "Children" is a word belonging to the group "person", this input sentence is determined to match this Japanese pattern. The predicate ``Genkida'' is not included in the adjective clause, so the English pattern is ``《人》be fine.
” is selected.

After that, the conversion generation unit 4 converts "Children" from Japanese to English, changing "be" to "is, are,"
After performing morpheme adjustment such as "was, were", etc., the translated sentence "A child is fine" is finally obtained.

[0017] If the Japanese sentence to be input next includes the expression ``a cheerful child,'' the valence pattern dictionary 7 is searched using the predicate ``genki da'' as a key, and the Japanese sentence ``《人》 ga ki dan'' is searched. Get the pattern. This time, the question is whether "child" modified by "genki" is related to "genki" or not, that is, the expression "kid *genki da (* is a particle such as ga/wo/ni)" is established. It is first determined whether or not to do so. “Child” is 《
Since the child belongs to the ``person'' group, it is judged to match the Japanese pattern for ``the child is doing well.'' The predicate ``Genkida'' is included in an adjective clause, and it incorporates the modifying noun ``Child'' as a case element, so the English pattern is ``《
"a child" is selected.The conversion generator 4 once selects "a child who is he".
When the relative clause "who" is degenerated, the English expression "a healthy child" is obtained.

[0018] Furthermore, if the input Japanese sentence includes the expression "a child whose parents are energetic," the valence pattern dictionary 7 is searched using the predicate "genki da" as a key, and the phrase "<person> is energetic" is searched. Get Japanese pattern. ``Energetic'' is a predicate of an adjective clause, but ``parent'' included in the adjective clause is a ``person''.
Since it belongs to the group ``My parents are fine'' is determined to match the Japanese pattern. The predicate ``Genkida'' is included in the adjective clause, but the noun ``Child'' that it modifies was not included as a case element, so the English pattern is ``《
"parents are fine" is selected in the conversion generation unit 4, and "parents are fine" is selected.
If a Japanese → English translation system can determine the relationship between “child” and “parent,” then “a child who
You can get English expressions like "parents are fine".

FIG. 3 shows a flowchart of the operation of the pattern converter according to an embodiment of the present invention. The following flowchart shows the processing after a Japanese sentence is input from the input unit 1, and data subjected to word analysis and syntactic analysis is supplied to the language analysis unit 2 using the analysis dictionary 6.

Step 31: Extract predicates from the input Japanese sentence.

Step 32: Search the bond valence pattern dictionary 7 using the predicate as a key.

Step 33: It is determined whether there is a Japanese pattern corresponding to the key in the bond valence pattern dictionary 7, and if there is no pattern, the process moves to step 41.

Step 34: Japanese patterns are taken out in sequence from the valence pattern dictionary 7 and compared with the analysis results of the language analysis section 2.

Step 35: It is determined whether the analysis result matches the Japanese pattern, and if it does not match, the process moves to step 40.

Step 36: If the analysis result matches the Japanese pattern, it is determined whether the predicate of the analysis result is included in an adjective clause and whether the modified noun is a case element. If this selection condition is not met, the process moves to step 39.

Step 37: When the selection condition of step 36 is satisfied, it is determined whether there is an English pattern for an adjective clause in the valence pattern dictionary 7. If the pattern exists, the process moves to step 38; otherwise, the process moves to step 39.

Step 38: If there is an English pattern for an adjective clause in the valence pattern dictionary 7, that pattern is selected and passed to the conversion generator 4.

Step 39: If there is no English pattern for adjective clauses in the valence pattern dictionary 7, a general English pattern is selected and passed to the conversion generator 4.

Step 40: If the analysis result does not match the Japanese pattern, step 34 is executed until there are no more matching patterns, but when the matching pattern is finished, the process moves to step 41.

Step 41: The conversion generator 4 converts each word from Japanese to English.

Step 42: Generate the converted English as an English sentence and send it to the output unit 5.

The pattern conversion result obtained in step 39 of the above flowchart is arranged into a natural sentence by the conversion generation section 4 using the conversion generation dictionary 8, and outputted by the output section 5.

[0033] The above description has been made on the case where the input Japanese sentence matches the Japanese pattern, in which the effects of the present invention occur. However, when the input Japanese sentence does not match the Japanese pattern, the conversion generation dictionary 8
For example, by registering an English translation for ``I'm fine'' in , a translated sentence will be generated by word replacement.

[0034]

[Effects of the Invention] According to the present invention, for predicates that are translated differently depending on whether they are used as a normal sentence or as an adjective clause, the pattern can be registered in the bond valence pattern. It is possible to provide translations suitable for the target language.

Furthermore, since the source language pattern in the valence pattern dictionary is shared between normal predicates and adjective clause predicates, it does not affect the efficiency of source language analysis. Furthermore, other than selecting a target language pattern by paying attention to whether it is an adjective clause or not during pattern matching, no new functions are required when converting and generating the target language, so it is efficient.

[Brief explanation of the drawing]

FIG. 1 is a basic block diagram of one embodiment of the present invention.

FIG. 2 is a diagram showing the structure of a bond valence pattern dictionary according to an embodiment of the present invention.

FIG. 3 is a flowchart of the operation of a pattern converter according to an embodiment of the present invention.

[Explanation of symbols]

1 Input section 2 Language analysis department 3 Pattern conversion section 4 Conversion generation unit 5 Output section 6 Analysis dictionary 7. Bond valence pattern dictionary 8 Conversion generation dictionary

Claims (1)

[Claims] Claim 1: A plurality of expression patterns of a second natural language and their selection conditions are described for expression patterns of case elements and predicates in a first natural language, using the word of the predicate as a key. a bond valence pattern dictionary, and the first
an input means for reading an input sentence that is a natural language sentence; a language analysis means for word segmentation and syntactic analysis of the input sentence read by the input means; and a valence pattern dictionary based on the linguistic analysis results. , select the predicate of the input sentence from the plurality of patterns of the second natural language having a plurality of expression patterns, and if there is one that matches the selection condition, select that expression pattern to obtain a pattern conversion result. a pattern conversion means having a function, and an output means for outputting a sentence in the second natural language after translating words in the first natural language into the second natural language based on the pattern conversion result; Using a natural language natural translation device comprising A second natural language expression pattern is registered in the valence pattern dictionary, and the pattern conversion means determines whether the predicate in the input sentence is a normal predicate or an adjective clause predicate. A natural language automatic translation method characterized by selecting an expression pattern in addition to the selection conditions of the selection function.