JPH05290084A - Unregstered word translation system - Google Patents

Abstract

PURPOSE:To synthesize an appropriate translated word at the time of translating an unregistered word into Japanese. CONSTITUTION:When an input sentence is inputted from an input part 1, an analysis part 2 refers to a dictionary 6 and a grammar rule 7 and analyzes the structure of the input sentence. When there is the unregistered word in the input sentence at that time, the analysis part 2 estimates the part of speech and analysis is continued. When the unregistered word is estimated to be a noun, the analysis part 2 decides the meaning origin and the usage of the unregistered word. Namely, the analysis part 2 estimates an attribute and the like which are required for translating the unregistered word. Then, a conversion generation part 3 generates a Japanese sentence based on an analyzed result. When the unregistered word is the noun, the corresponding translated word is synthesized. Thus, the conversion generation part 3 generates the Japanese sentence by using the synthetic translated word.

Description

Detailed Description of the Invention

[0001]

TECHNICAL FIELD The present invention relates to an unregistered word translation system, and more particularly, to a machine translation device, and more particularly to an unregistered word translation system in an English-Japanese machine translation device. For example, it is applied to a translation device that translates a natural language using a dictionary.

[0002]

2. Description of the Related Art In a machine translation device using a dictionary, it is impossible to register all the words appearing in an input sentence in the dictionary in advance, and it is inevitable that unregistered words will occur. Therefore, in order to be able to process even sentences containing unregistered words, the attributes of unregistered words are estimated from the attributes expected from the syntax analysis rules, and the analysis is performed using those attributes. An unregistered word processing method for attribute estimation with higher performance has been proposed, and it has become possible to analyze even sentences containing unregistered words.

The syntax analysis method is proposed in, for example, Japanese Patent Laid-Open No. 175075/1983. In this publication, the attribute of a dictionary unregistered word is estimated by the attribute expected from the syntax analysis rule. Also,
The unregistered word processing method is, for example, Japanese Patent Laid-Open No. 63-
It is proposed in Japanese Patent No. 68972, and an attribute is estimated from information obtained by decomposing a dictionary unregistered word into a prefix, a suffix, and a word base.

However, regarding the translation processing of unregistered words,
The unregistered word is displayed as the original word, or the translation word is synthesized by combining the affix and the translation word of the word base in a fixed form, and the synthesized translation word is inappropriate as the target language word. Many things happened. For example, in the English-Japanese machine translation device, the sentence (1) below is input, and of these, "analyse
It is assumed that "r" is not registered in the dictionary, but the base "analyse" is registered. (1) "This system is equipped with an automatic
“Analyser.” In the conventional technique for performing a fixed word synthesis, for example, “anal
The translation of “analyser” is synthesized by combining the translation of “yse” “analyze” and the translation of “er” “person”.
You will get an output like (2) “This system is equipped with a person who analyzes automatically.” However, in this case, “analyser” means a device, not a person, so “analyzer” is an inappropriate translation. Is. Furthermore, it is a poorly translated word that is redundant even in Japanese, and it is not possible with the conventional technology to synthesize and translate appropriate translated words such as "analyzer" and "analyzer". Further, in the conventional technique in which an unregistered word is displayed as it is, the output as shown in (3) below is obtained. (3) “This system is equipped with an automatic analyzer.” In this method, even though the information on “analyse” is in the dictionary, the information is not reflected in the translated text, so the user can “analyse”. If you don't know the meaning of "," you need to look up the dictionary.

[0005]

It is an object of the present invention to provide an unregistered word translation system that synthesizes appropriate translated words when translating unregistered words into Japanese. It was done for the purpose.

[0006]

In order to achieve the above object, the present invention provides (1)
For the source language words and affixes, information such as their attributes and case patterns, a dictionary that describes Japanese translations, and a dictionary that is not registered in the dictionary but is a derivative form of the words registered in the dictionary, and Estimating means for estimating the semantic feature of an unregistered word that is presumed to be a syntactic noun by using the case pattern of co-occurring verbs, and the translated word of the unregistered word, which is estimated by the estimating means. And (2) the unregistered word is a derivative noun from a verb registered in the dictionary, and (2) the unregistered word is derived from the translated word of the word registered in the dictionary according to the semantic feature of the word. Sometimes, a translated word of the verb described in the dictionary is used as a noun to synthesize a translated word of the unregistered word, and further, (3) in (2), the unregistered word is When the Japanese translation of the constituent verb is a sa verb, The word stem is used to synthesize a translated word of the unregistered word, or (4) information about attributes and case patterns of words and affixes in the source language, a dictionary describing Japanese translated words, and the dictionary Although it is not registered in, the semantic features of unregistered words that are derived from verbs registered in the dictionary and are presumed to be syntactically nouns Of the translation means of the verb registered in the dictionary, which is estimated using the semantic feature, the translation word of the unregistered word is converted into a noun by converting the translation word of the verb corresponding to the semantic feature estimated by the estimation means. And a synthesizing means for synthesizing. Hereinafter, description will be given based on examples of the present invention.

FIG. 1 is a block diagram for explaining an embodiment of an unregistered word translation system according to the present invention. In the figure, 1 is an input unit, 2 is an analysis unit, 3 is a conversion generation unit, and 4 is an output. Part, 5 is an affix dictionary, 6 is a dictionary part, 7 is a grammar rule, 8 is an input sentence, and 9 is an output sentence. Attribute information such as case patterns and semantic features and its Japanese translation are described in the dictionary 6 for words in the source language, for example, as shown in FIG. 2, and this information is added to normal translation processing. Also used for translation processing of unregistered words.
Similarly, as shown in FIG. 3, the affix dictionary 5 describes the part-of-speech attribute of the word base to be connected, the translation of the affix, and the like.

When an input sentence is input from the input unit 1, the analyzing unit 2 refers to the dictionary 6 and the grammar rule 7 to analyze the structure of the input sentence. At this time, if there is an unregistered word in the input sentence, the analysis unit 2 estimates the part of speech and continues the analysis. As the method of estimating the part of speech, there is the method described in the above-mentioned Japanese Patent Laid-Open No. 63-68972. Now, when the unregistered word is estimated to be a noun by this, the analysis unit 2 determines the semantic feature and usage of the unregistered word by performing the process shown in FIG. Hereinafter, description will be given based on each step.

Step 1 ; It is judged whether or not the unregistered word can be decomposed into a word or an affix registered in the dictionary, and if it cannot be decomposed, the process ends. step2 ; If the unregistered word consists of a word and an affix, it is judged whether or not the suffix has a plurality of translated words. If the suffix has a plurality of translated words, the process proceeds to step3, and if not, the process proceeds to step4. step3 ; The semantic feature of the unregistered word is determined by referring to the case pattern of the verb that co-occurs with the unregistered word. step4 ; It is judged whether or not the word which is the base of the unregistered word has a plurality of usages (translated words), and if there is only a single usage (translated word), the processing is ended. step5 ; It is judged whether the word base of the unregistered word is a verb, and if it is not a verb, the processing ends. step6 ; It is judged whether or not the unregistered word constitutes a compound noun together with other dictionary registered nouns, and if not, the processing ends. step7 : Determine the usage of the word base verb by using the co-occurrence description of the dictionary about the word base verb of the unregistered word and the semantic feature of the noun co-occurring.

By the above processing, the analysis unit 2 estimates the attributes and the like necessary for translating unregistered words. As an example, when the English sentence (1) is input, "analyser" is presumed to be a noun, and by the process shown in FIG. 4, from the dictionary description of the co-occurring verb "be equipped with" [TOO
It is presumed to have an attribute of L].

Next, based on the analysis result in the conversion generator 3,
Words are generated. If the unregistered word is a noun here,
The corresponding translated word is processed according to the process shown in the flow of FIG.
Is made. Hereinafter, description will be given based on each step.step1 ; Unregistered words are composed of dictionary registered words and affixes
If yes, go to step 11.
Ku.step2 ; Determine if the unregistered word contains a verb, and if not
Go to step 7.step3 ; Determine if the verb is a simple phrase
If so, go to step 7.step4 ; Determine if the verb is a sa-verb and so
If not, go to step6. step5 Use the stem of the verb's translation for translation synthesis.step6 ; If the noun form of the verb's translation is in the dictionary, translate it
Used for synthesis.step7 ; Determine if there are multiple translations of the dictionary registration word
If there is more than one, go to step 8, otherwise go to step 9.
Ku.step8 ; Translated words into one by using attributes estimated during analysis
Determine if squeezed, otherwise go to step 10.
Ku.step9 ; Translation of unregistered words divided into dictionaries (verb
Regarding translated words, those translated into nouns in steps 2 to 7)
And synthesize the translated words.step10 If multiple translations are not possible, combine multiple translations
To achieve.step11 ; For unregistered words that do not include dictionary registered words,
Translate the original language.

As an example, since the above "analyser" is presumed to be a noun, it is judged in step 1 whether the word can be divided into dictionary registered words. In this case, since it is known that the dictionary registration verb "analyse" and the suffix "er" can be divided, the process proceeds to step 2. Since the verb is included, st
Proceeding to ep3, since the translated word "analyse""analyse" is a syllable composed of a single phrase, the stem "analyze" is generated as a translated word in step 5. Then go to step 7.

Although there is only one translated word for "analyse", "er" has two translated words (usage), "person" and "apparatus".
Go to ep8. By the attribute [TOOL] estimated by the analysis unit,
Since the translated word of "device" is selected, go to step 9 and combine "analysis" and "device", and the unregistered word "analyser"
“Analysis device” is synthesized as a translation of. In this way, the conversion generating unit generates a Japanese sentence using the translated words synthesized.

As described above, the present invention estimates the part-of-speech and semantic feature attribute of an unregistered word, and when the unregistered word is a derivative of the dictionary registered word, analyzes the dictionary registered word from the dictionary. Means, and a translation word synthesizing means for synthesizing a translation word suitable for an unregistered word by using the attribute and dictionary information estimated by the analysis means, and when the unregistered word is a noun, its semantic feature Estimate by referring to the case pattern of the verb that co-occurs with, and perform translation based on its feature. At that time, when the unregistered word is a derivative of the verb of the dictionary registered word, the translated word of the verb Is used to synthesize a translation of an unregistered word, and when the translation of the verb is a sa verb, the stem of the translation is automatically adopted as a nounized expression of the verb. Synthesizing translations of unregistered words, and When the registered word is a derivative noun from the verb of the dictionary registered word and is a compound noun that co-occurs with a noun of another dictionary registered word, the case pattern of the verb or the co-occurrence description, The feature is that, by using the attribute information of the co-occurring noun, the translation of the unregistered word is synthesized using the most suitable translation of the verb.

[0015]

As is apparent from the above description, the present invention has the following effects. (1) In the unregistered word translation method of claim 1, the attribute of the unregistered word is estimated using the case pattern of the co-occurring verb,
Since the translated word of the unregistered word is synthesized from the translated word of the dictionary registered word according to the attribute, the unregistered word can be appropriately translated. (2) In the unregistered word translation method according to claim 2, since the translated word of the verb that constitutes the unregistered word is converted into a noun for translation of the unregistered word, more natural translation can be performed. (3) In the unregistered word translation method of claim 3, when the translated word of the verb forming the unregistered word is a sahen noun, the stem is used for translating the unregistered word, so that a more natural translation is possible. (4) In the unregistered word translation method according to claim 4, when a dictionary registered word having a plurality of translated words (usages) constitutes an unregistered word, if the word is a verb, it is regarded as another co-occurring noun. Since the translated word (usage) is determined from the case relation and used for translating an unregistered word, it can be translated appropriately.

[Brief description of drawings]

FIG. 1 is a configuration diagram for explaining an embodiment of an unregistered word translation system according to the present invention.

FIG. 2 is a diagram showing information in a dictionary section of the present invention.

FIG. 3 is a diagram showing information of an affix dictionary according to the present invention.

FIG. 4 is a flowchart showing the operation of the analysis unit of the present invention.

FIG. 5 is a flowchart showing the operation of the conversion generation unit of the present invention.

[Explanation of symbols]

1 ... input part, 2 ... analysis part, 3 ... conversion generation part, 4 ... output part, 5 ... affix dictionary, 6 ... dictionary part, 7 ... grammar rules, 8 ... input sentence, 9 ... output sentence.

Claims (4)

[Claims] 1. For a word or affix in the source language, information such as its attribute or case pattern, a dictionary describing Japanese translations, and a word that is not registered in the dictionary but is registered in the dictionary An estimator that estimates the semantic feature of an unregistered word that is a derivational form and is presumed to be a syntactic noun using a case pattern of co-occurring verbs, and a translated word of the unregistered word by the estimator. An unregistered word translation method comprising: a synthesizing unit that synthesizes from translated words of words registered in a dictionary according to the estimated semantic features of the unregistered words. 2. When the unregistered word is a derivative noun from a verb registered in the dictionary, the translated word of the verb described in the dictionary is converted into a noun to use the unregistered word of the unregistered word. The unregistered word translation method according to claim 1, wherein translated words are synthesized. 3. The non-registered word according to claim 2, wherein when the Japanese translation of the verb forming the unregistered word is a sa verb, the stem of the verb is used to synthesize the translation of the unregistered word. Registered word translation method. 4. For a word or affix in the source language, information such as its attribute or case pattern, a dictionary describing Japanese translations, and a verb that is not registered in the dictionary but is registered in the dictionary Registered in the dictionary, an estimating means for estimating the semantic feature of an unregistered word that is a derivational form and is presumed to be a syntactic noun using the semantic feature of the noun that co-occurs with the case pattern of the verb. Of the translated words of the verb, the word corresponding to the semantic feature estimated by the estimating means is converted into a noun,
An unregistered word translation method comprising: a synthesizing unit that synthesizes a translated word of the unregistered word.