JPH03214355A - machine translation device - 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 <Industrial Application Field> The present invention relates to a machine translation device that can generate natural translated sentences.

BACKGROUND ART Conventionally, there are the following mw translation devices. This machine translation device inputs a source language input from a keyhole into a translation module under the control of the main CPU, and converts the input source language into a dictionary consisting of a main dictionary and a user dictionary stored in memory, and a grammar. A translation module uses rules and tree structure conversion rules to translate into the target language.

When translating from English to Japanese using a machine translation device such as the one described above, when registering a new word in a user dictionary created by the user among the above dictionaries, it is necessary to input the following information: There is.

■ Heading of English words 2 Part of speech of English words 3. Inflectional forms of English words 4 Translations of English words 5 Parts of speech of translated words 6 Conjugation forms of translated words In this way, not only the 91 English words and their translations, but also the parts of speech of English words and translations. By registering the inflectional forms of English words and the conjugated forms of translated words in a dictionary, it is possible to obtain more natural translated sentences.

<Invention or problem to be solved> In this way, the English text input above is translated into Japanese text.
When registering a new TJ word in the user dictionary of the i-machine translation device, it is necessary to input the conjugated form of the translated word of the registered English word. Therefore, the user is required to have knowledge of Japanese grammar, and there is a problem in that not everyone can easily and efficiently create a user dictionary.

SUMMARY OF THE INVENTION An object of the present invention is to provide a machine translation device that allows anyone to easily and efficiently create a user dictionary.

<Means for Solving the Problems> In order to achieve the above object, the present invention provides a morphological analysis unit that divides an input sentence into morphemes and obtains the translation of the morphemes using a dictionary including a user dictionary. and a syntactic analysis unit that analyzes the syntax of the morpheme sequence divided by the morphological analysis unit using a dictionary and grammar rules, and converts the syntactic structure obtained by the syntactic analysis unit into a Japanese syntactic structure. In a machine translation device having a conversion unit and a translated sentence generation unit that generates a translated sentence in Japanese according to the Japanese syntactic structure obtained by the conversion unit, a Japanese translation corresponding to the word registered in the user dictionary is used. a translated word character determining means for determining whether or not a character at a predetermined position in a character string constituting the character string is a prespecified character; and a conjugation form of the translated word according to a predetermined rule based on the determination result by the translated word character determining means The present invention is characterized by comprising translated word inflection estimation/registration means for automatically estimating and registering the estimation result in the user dictionary.

Effects> When a sentence is manually input to the morphological analysis unit, the input sentence is divided into morpheme sequences by the morphological analysis unit to obtain Japanese translations and the like. When the input sentence is divided into morpheme sequences in this manner, the syntax of the upper occurrence morpheme sequences is analyzed by the syntax analysis section. Then, the syntactic structure obtained by the syntactic analysis section is converted into a Japanese syntactic structure by the conversion section, and a translated sentence in Japanese is generated by the translation sentence generator according to the obtained Japanese syntactic structure.

The dictionary used during morphological analysis in the morphological analysis section and syntactic analysis in the syntactic analysis section mentioned above is the Japanese language corresponding to the word to be registered in the user dictionary when registering a new word in the user dictionary. The translated word is input. Then, the translated word character determining means determines whether a character at a predetermined position in the character string constituting the translated word is a predetermined character. Then, based on the discrimination results by the above-mentioned translated word character discrimination means, the translated word inflection is estimated and
The registration means estimates the inflected form of the translated word according to a predetermined rule, and registers the obtained estimation result in the user dictionary.

Therefore, when a user registers a new word in the user dictionary, the conjugated form of the Japanese translation of the registered word is automatically estimated and registered in the user dictionary.

<Example> Hereinafter, the present invention will be explained in detail with reference to illustrated embodiments.

FIG. 1 is a block diagram of an embodiment of a machine translation apparatus according to the present invention. 1 is the main CPU (central processing unit),
2 is a main memory, 3 is a CRT (cathode ray tube), 4 is a keyboard, 5 is a translation module, and 6 is a translation dictionary consisting of a main dictionary, a user dictionary, etc. This is a memory that stores grammar rules, tree structure conversion rules, etc.

When a sentence in the source language is input, the translation module 5 translates it and outputs a sentence in Japanese, which is the target language. That is, Keyhot 4
The input source language is sent to the translation module 5 under the control of the main CPUI. The translation module 5 is a dictionary stored in the memory 6. Using grammar fill, tree structure conversion rules, etc., the input source language is translated into the target language as will be detailed later. The results are temporarily stored in the main memory 2 and also
3.

Automatic translation performed in translation module 5 above. In general, there are analysis levels as shown in Figure 2. When the source language is input in the upper left of the figure, dictionary lookup in Lehel L1,
Morphological analysis at level L. Level L3 parsing...
・The analysis continues.

Machine translation can be broadly divided into the following two types depending on the level of analysis. One is to analyze concepts that do not depend on either the source language or the target language at level L6 (called an intermediate language), and from there, to generate context at level L7. This is a pivot method in which the target language is generated by proceeding with meaning generation at level L8, syntax generation at level Lll, and morpheme generation and generation at level L+o.

1-) includes the above-mentioned morphological analysis at level L, syntactic analysis at level L4, semantic analysis at level L4, and level L
, to obtain the internal structure of the source language, and then convert it to the internal structure of the target language at the same level as the internal structure of the source language obtained, and then generate the target language. This is a one-way transfer system.

Here, each of the above analyzes will be explained.

・Dictionary lookup・Morphological analysis The input sentence is divided into each morpheme sequence (single (word strings) and create the menu shown in Figure 1. Look up each word in Mori 6's dictionary. Grammatical information such as parts of speech and Obtain translations, and also learn tenses, personal names, Analyze numbers, etc.

・Syntactic analysis: Determine the structure of the sentence (structural analysis tree), such as the dependencies between words, as described later.

・Semantic analysis: Distinguish between what is semantically correct and what is not based on the results of multiple syntactic analysis.

・Context analysis: Understand the topic and remove omissions and ambiguities.

It is assumed that the translation module 5 used in this invention performs analysis up to at least level L3 structural analysis level. That is, the translation module 5 of this embodiment includes a dictionary lookup morphological analysis section ll. as shown in FIG. Let us consider a translation module consisting of the functional parts of the syntactic analysis section 12, the conversion section 13, and the translated text generation section 14.

FIG. 4 is a concrete block diagram of the translation module 5. Also, Figures 5 to 9 are rThis isape.
n. The fourth example when translating the English sentence J into Japanese.
The contents of each buffer A to E shown in the figure are illustrated.

The English-Japanese translation process will be explained below using FIGS. 3 to 9.

First, the read original text is stored in buffer A in FIG. 4, as shown in FIG. Then, under the control of the translation CPUI5 based on the translation program l6, the buffer A
Necessary information such as the translation of each word is obtained using the dictionary in the memory 6 according to the original text stored in the memory 6, and is stored in the buffer B (dictionary lookup and morphological analysis by the dictionary lookup morphological analysis unit 2 in FIG. 3). . As a result, part-of-speech information, which is part of the above information, is stored as shown in FIG. 6, for example. Here, rt
Although hisJ is a multipart speech word, it is uniquely determined by the following syntactic analysis. That is, according to the dictionary in the memory 6 and the grammar rules, a structural analysis tree showing the dependency relationships between each word is determined as shown in FIG. 7, and stored in the buffer C in FIG. (Syntax analysis by syntax analysis unit l2).

This structural analysis tree is determined as follows. That is, from the grammar rules in memory 6, sentence → subject, predicate subject → noun phrase predicate one verb, noun phrase noun phrase one pronoun noun one article. You can get rules for nouns. This rule, for example,
'llll means that "a sentence is made up of a subject and a predicate." Thereafter, a structural analysis tree is determined according to this rule.

Next, in the same way as in the case of syntax analysis described above, the tree structure conversion rule III in memory 6 is used to convert the structure of the parse tree (see Figure 7) for the input English sentence into a Japanese sentence as shown in Figure 8. is converted into a structure for . Then, the obtained result is stored in the bath sofa D in FIG. 4 (syntax conversion by the conversion unit 13 in FIG. 3). Finally, the obtained Japanese character ``This is Ben desu.'' is given the appropriate particle ``wa'' or an auxiliary verb.
It is stored in the buffer E of FIG. 4 in the Japanese form as shown in FIG. 9 (translated sentence generation by the translated sentence generation unit 14 of FIG. 3). The Japanese sentence thus obtained was ``This is Ben.''
is output from the translation module 5, stored in the main memory 2, and displayed on the CRT 3.

During the above-mentioned translation process, the user uses a pre-used main dictionary as a dictionary to be used for dictionary lookup and morphological analysis, as well as special words that are not in this main dictionary as needed. There are user dictionaries that can be registered and created. When registering a new English word and a Japanese translation corresponding to this English word in the user dictionary, the translation CPU 5 executes the user dictionary registration process.

FIG. 10 is a flowchart of the user dictionary registration processing operation performed by the translation CPU 15 of FIG. below,
The user dictionary registration processing operation will be described in detail with reference to FIG.

At step S1, the heading of an English word to be registered in the user dictionary is input and registered in the user dictionary.

In Step S2, the part of speech of the English word input in Step S1 is input and registered in the user dictionary.

In step S3, the inflectional form of the English word input in step S1 is input and registered in the user dictionary.

In step S4, the Japanese translation of the English word input in step S1 is input and registered in the user dictionary.

In step S5, the part of speech of the translated word input in step S4 is input and registered in the user dictionary.

In step S6, the conjugated form of the translated word manually inputted in step S4 is input and registered in the user dictionary, and the user dictionary registration processing operation is completed.

At that time, if the part of speech of the translated word input in step S5 is a verb, the translated word input in step S4 is input under the control of translation CPU15 when inputting the translated word conjugation form in step S6. The conjugated form of the translated word is automatically estimated based on the characters in the translated word.

FIG. 11 is a flowchart of the translated word inflection estimation/registration processing operation for verbs executed by the translation CPU 15. The translated word inflection estimation/registration processing operation will be described in detail below with reference to FIG.

In step S11, the character after D of the translated word registered in the user dictionary is read out and manually entered in step S4 in the flowchart of the user dictionary registration processing operation.

In step S12, it is determined whether the last character of the input translated word is "u/</gu/su/tsu/nu/fu/mu". As a result, “U/</</su/tsu/nu/
If the answer is either "F/M", the process advances to step S13; otherwise, the process advances to step Sl4.

Step S13 estimates that the conjugation form of the registered translated word is a five-stage conjugation, and this estimation result is stored in a buffer (not shown).

In step Sl4, the last character of the input f2 translated word is "
It is determined whether or not it is. If the result is "ru", the process proceeds to step SI5, and if not, the translated word inflection estimation/registration processing operation is ended.

In step S15, the second to last character of the translated word registered in the user dictionary is read and input.

In step S16, it is determined whether it is the second character from the end of the input translated word or "Ki/<". If the result is either "come/<", the process advances to step Sl7; otherwise, the process advances to step S18.

At Stenop S17, the conjugation form of the registered translated word is estimated to be a kinetic declension conjugation, and this estimation result is stored in the above-mentioned banofa.

At Stenop S18, it is determined whether the second to last character of the translated word that has been swallowed is either "su/zu". If the result is either "su" or "zu", the process proceeds to step SI9; otherwise, the process proceeds to step S20.

In step S19, the conjugation form of the registered translated word is estimated to be the sagyohenban conjugation, and this estimation result is stored in the buffer.

In step S20, the penultimate character of the human translated word is [ke/ge/什/ze/te/te/ne/he/be/pe/
It is determined whether it is either "me" or "me".

As a result, “ke/ge/se/ze/te/de/ne/he/be/pe/
If it is either "M/R", the process advances to step S21; otherwise, the process advances to step S22.

In step S21, the conjugation form of the registered translated word is estimated to be the lower 1st conjugation, and this estimation result is stored in the buffer.

In step S22, the inflection form of the registered translated word is estimated to be one of the five-dan conjugation, the upper one-dan conjugation, and the lower one-dan conjugation, and the CRT 3 prompts the user to select the correct inflection.

In step S23, the five-stage utilization estimated in step S22. Utilize the upper level. The user selects which inflection form among the lower one-stage inflections, and the selected inflection form is stored in the buffer.

In step S24, the inflections estimated in each of the above steps and stored in the buffer are registered in the user dictionary of the memory 6, and the translated word inflection estimation/registration processing operation is completed.

English word headings in the user dictionary registration process described above.
Input of part of speech, inflectional form, translated word, and translated part of speech and selection of conjugated form in the translated word conjugated form estimation/registration process are performed manually by inputting the necessary information from the keyboard 4 according to the instructions displayed on the CRT 3 under the control of the translation CPU 15. It is carried out by

In this way, in this embodiment, when registering a new word in the user dictionary used when translating an input English sentence into a Japanese sentence, if the input Japanese translation is a verb or a verb, the last word of the translated word is The conjugated form of the above translated word is automatically estimated based on the character and the second-to-last character and registered in the user dictionary, so even users without knowledge of Japanese grammar can easily use it. This makes it possible to efficiently execute translated word conjugation input processing.

In the above embodiment, the translated word inflection estimation/registration processing operation is explained using the estimation of the inflection of a verb as an example, but there is no problem in estimating the inflection of other parts of speech.

The algorithm for estimating translated conjugation forms for verbs in this invention is not limited to the above embodiments.

Further, the configuration and contents of each bath sofa of the translation module 5 in this invention are not limited to the present embodiment.

Although the source language in the above embodiment is English, it goes without saying that the present invention is not limited thereto.

<Effects of the Invention> As is clear from the above, the machine translation device of the present invention has the following effects:
Equipped with a translated character discrimination means and a translated word inflection estimation/registration means, when registering a new word in the user dictionary, the characters at a predetermined position in the character string constituting the Japanese translation corresponding to the registered word are specified in advance. The translated word character discriminating means determines whether the translated word is a translated character, and the translated word inflection estimation/registration means automatically determines the translated word's inflected form according to a predetermined rule based on the determination result by the translated word character discriminating means. Since the estimation result is registered in the user dictionary above, when a new word is registered in the user dictionary, if the user inputs the Japanese translation corresponding to this registered word, it will be automatically The conjugated form of this translated word is estimated and registered.

Therefore, according to the present invention, even a user who has no knowledge of Japanese grammar can easily and efficiently register the conjugated form of a translated word.

[Brief explanation of drawings]

Fig. 1 is a block diagram of an embodiment of the machine translation device of the present invention, Fig. 2 is an explanatory diagram of the translation level in automatic translation, Fig. 3 (voice) is a functional partial configuration diagram of the translation module in the above embodiment, and Fig. 4 The figure is a Bronnock diagram of the above translation module, Figures 5 to 9 are diagrams showing examples of the contents of each buffer 7 in Figure 4, Figure 10 is a flowchart of user dictionary registration processing operation, and Figure 11 is a translated word inflection form. This is a flowchart of estimation/registration processing operation. l Main CPTJ, 2...Main memory, 3
CR T, 4. Keyword, 5. Translation monoule, 6. Memory, II.. Dictionary lookup morphological analysis unit, l2 Syntactic analysis unit, I3. Conversion unit,
I4・Translation sentence generation unit, l5...Translation CPU, 1
6. Translation program, A-E. Buffer. 5y example sentence fThis pen. 》 @；6[2l

Claims (1)

[Claims] (1) A morphological analysis unit that divides an input sentence into morphemes and uses a dictionary including a user dictionary to obtain translations of the morphemes, and a morphological analysis unit that analyzes the syntax of the morpheme string divided by the morphological analysis unit using a dictionary and grammar rules. a syntax analysis unit that analyzes using
A conversion unit that converts the syntactic structure obtained by the above syntactic analysis unit into a Japanese syntactic structure, and a translated sentence generation unit that generates a translated sentence in Japanese according to the Japanese syntactic structure obtained by the above conversion unit. In a machine translation device having
a translated character determining means for determining whether or not a character at a predetermined position in a character string constituting a Japanese translation corresponding to a word registered in the user dictionary is a predetermined character; A machine translation device comprising translated word inflection estimation/registration means for automatically estimating the inflected form of the translated word according to a predetermined rule based on the determination result and registering the estimation result in the user dictionary.