JP3287068B2 - Natural language automatic translator - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language automatic translator for automatically translating a sentence in a natural language (source language) into a sentence in another natural language (target language).

[0002]

2. Description of the Related Art As a conventional natural language translation technique, a technique based on the concept of a so-called valency pattern is known, as described in, for example, Japanese Patent Application Laid-Open No. 63-86071. In this technique, a source language sentence is first analyzed into a tree structure described in a dependency relationship. Next, the relation between the predicate corresponding to the root and the clause leading to the predicate is extracted. Furthermore, these phrases are compared with the source language expression pattern. A corresponding target language expression pattern is prepared for the matched source language expression pattern, and a target language sentence is generated based on the target language expression pattern.

[0003]

In the above prior art,
Only the predicate and the clause that directly leads to the predicate are matched against the source language expression pattern. Therefore, when a sentence that is not directly connected to a predicate in a source language sentence is translated into a sentence that is essential in source language sentence analysis and translation, the relationship between the same phrase and the predicate is not correctly analyzed, and an appropriate target language sentence is generated Not be
Manual correction was required. For example, in order to translate “I develop the system” to “I developsystem.”, The three people “I”, “system”, and “development” must be in the relationship of “I develop the system”. It is necessary to capture something, but with the conventional valency pattern technology, the relationship between "I", "development" and "do", and the relationship between "system" and "development" were captured independently. ,
For example, a translation such as "I dodevelopment of a system" was output.

[0004] The present invention has been made in view of the above,
It is an object of the present invention to provide a natural language automatic translation apparatus that correctly analyzes and translates a phrase by using an expression pattern that describes a clause that is not directly connected to a predicate by a tree structure of three or more stages.

[0005]

In order to achieve the above object, the present invention provides an analyzing unit for analyzing a source language to be translated into a source language structure described in dependency relations, and an analysis unit based on the source language structure. A natural language automatic translation apparatus having a conversion unit for converting the expression pattern of the source language into an expression pattern of the target language, and a generation unit for generating a translation target language based on the target language expression pattern. An expression bilingual dictionary in which an expression pattern of a source language described in a tree structure described above and an expression pattern of a target language corresponding to the expression pattern of the source language are stored in advance, and the conversion unit includes three or more levels of tree For the source language structure described by the structure, the expression pattern of the source language is converted into the expression pattern of the target language using the expression bilingual dictionary.

Specifically, the expression pattern of the source language and the source
The target language expression pattern corresponding to the language expression pattern
In the bilingual expression dictionary that stores
Each component of the source language structure is defined as a predicate as a key word.
Has a dependency condition, and the condition includes a predicate.
Take a tree structure of three or more levels at a point, and connect directly to the predicate.
For source language structures that contain
Includes the dependency conditions of components that do not directly
Confuse. The conversion unit searches the source language structure obtained by the analysis unit for an expression pattern of the source language that satisfies the dependency condition of each component of the source language structure from the expression bilingual dictionary using the predicate as a key. An expression pattern of the target language corresponding to the expression pattern of the source language is obtained.

[0007]

According to the automatic natural language translator of the present invention, a predicate is
-Elements that are words and are not directly connected to predicates in the source language
By using an expression bilingual dictionary in which expression patterns of a source language sentence in which dependency conditions between components including elements (phrases) are described and expression patterns of a target language corresponding to the expression patterns of the source language sentences are stored in advance. , Directly connected to the predicate
Translating sentences that are indispensable in source language sentence analysis
In this case, a target language sentence having an appropriate structure can be translated.

[0008]

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a schematic diagram showing an embodiment of a natural language automatic translation apparatus according to the present invention. The source language sentence input from the input unit 1 is analyzed by the analysis unit 2 into a tree structure described in a dependency relationship with reference to the analysis dictionary 6, and then the translation unit 3 refers to the expression bilingual dictionary 7. To the target language expression pattern. The features of the present invention reside in the conversion unit 3 and the expression bilingual dictionary 7, which will be described later in detail. Subsequently, based on the converted language expression pattern, the generation unit 4 obtains the target language sentence by referring to the generation dictionary 8, and the output unit 5 outputs the target language sentence.

FIG. 2 is a diagram showing a specific configuration example of the expression bilingual dictionary 7 in FIG. Here, the case where the source language is Japanese and the target language is English is taken up. In the bilingual expression dictionary 7, a Japanese expression pattern 71 and a corresponding English expression pattern 74 are described. Furthermore, Japanese expression pattern 7
1 includes a word (predicate) 75 serving as a search key for the pattern.
And a condition 73 for each component 72 of the Japanese expression. Here, for example, in pattern number 100,
Component D is directly linked to the key word (predicate) of “do”
It is a component that does not. Similarly, for pattern number 110
The component D is directly connected to the key word “do”
It is a component that does not. In the English expression pattern 74,
For each key word 75, an English expression structure to be generated based on the condition 73 of each component 72 is described.

The expression pattern of English (target language) for one key word is not limited to one, but may be plural. In that case, a pattern number may be provided for each expression pattern to distinguish them. For example, in addition to the example of FIG.
When an English expression pattern generated under conditions such as F exists, for example, a pattern number “101” is provided, the key word 75 is “do”, which is the same as the pattern number “100”, and the constituent elements 72 are “A”, The condition 73 may be described for each of the components as “E” and “F”, and the corresponding English expression structure may be described in the English expression pattern 74.

FIG. 3 is a flowchart showing the operation of the conversion unit 3 in FIG. The conversion unit 3 first extracts one pattern from the expression bilingual dictionary 7 (step 210).
If the pattern to be converted has not been subjected to the matching process in the dictionary 7 (NO in step 220), the sentence is converted to the target language structure using the conventional technique (step 28).
0). If there is a pattern (YES in step 220),
The input sentence is searched for a word that matches the key word described in the extracted pattern (step 230). If there is no key word in the input sentence (NO in step 240), the process returns to step 210 to extract the next pattern from the expression bilingual dictionary 7. If there is a key word (YE in step 240)
S) For each component of the Japanese expression pattern, a sentence element satisfying the condition is searched from the input sentence (step 250). If all the constituent elements are satisfied (YES in step 260), the corresponding target language expression pattern in the expression bilingual dictionary 7 is extracted (step 27).
0). On the other hand, if not satisfied (N in step 260)
O) Returning to step 210, the same processing is performed for the next pattern.

Next, in the case where the source language is Japanese and the target language is English, the processing flow of FIG. 3 will be concretely described with an example where the input sentence "I will develop the system" is translated. Will be described. The contents of the expression bilingual dictionary 7 are shown in FIG.

The input sentence is analyzed by the analysis unit 2 into a tree structure shown in FIG. FIG. 4 shows that "I" and "development" are connected to "do" at the root, and "system" is connected to "development".

First, in step 210, the pattern of pattern number 100 is extracted from the bilingual expression dictionary 7. Since the condition of step 220 is satisfied, the process proceeds to step 230. In step 230, "do" of the key word 75 is searched from the tree structure in FIG. Since “do” exists at the root of the tree structure in FIG. 4, the condition of step 240 is satisfied, and the process proceeds to step 250. In step 250, it is checked whether the condition 73 of each component 72 in the Japanese expression pattern 71 of the pattern number 100 is satisfied for the tree structure in FIG. In the example of FIG. 4, “do” is the condition of component A, “I am” is the condition of component B, “develop” is the condition of component C, and “system” is component D. Respectively. Therefore, the next step 260
Is YES, and the English expression pattern 74 having the pattern number 100 is extracted from the expression bilingual dictionary 7 at step 270 and passed to the generation unit 4. The generation unit 4 refers to the generation dictionary 8 and outputs, for example, “I develop a system.” As a result.

In the above output sentence, it is necessary for translation to capture the relationship between "I", "system", and "development", and in the prior art, the relationship between "I", "development", and "do", And "I doa development of a syst" which is expected when the relationship between "system" and "development" is considered independently.
em. This is better than "".

Next, the case where the input sentence is "I feel like going to lunch" will be described with reference to FIGS.

The input sentence is analyzed by the analysis unit 2 into a tree structure shown in FIG. First, in step 210, the pattern of pattern number 100 is extracted from the bilingual expression dictionary 7. Step 2
Since the condition of No. 20 is satisfied, the routine proceeds to step 230. In step 230, the key word "do" is retrieved from the tree structure in FIG. In FIG. 5, since “do” does not exist, the condition of step 240 is not satisfied, and the process returns to step 210. In step 210, the pattern of the next pattern number 110 is extracted from the bilingual expression dictionary 7. Since the condition of step 220 is satisfied, the process proceeds to step 230. In step 230,
The key word “do” is searched from the tree structure in FIG. Since “do” exists at the root of the tree structure in FIG. 5, the condition of step 240 is satisfied this time. In step 250, FIG.
It is checked whether or not the condition 73 of each component 72 in the Japanese expression pattern 71 of the pattern number 110 is satisfied for the tree structure of. In the case of the example of FIG. 5, “Yes” is the condition of the component A, “I am” is the condition of the component B,
“Ki ga” satisfies the condition of the component C, and “I want to go to lunch” satisfies the condition of the component D. Therefore,
The determination in step 260 is YES, and step 270
Then, the English expression pattern 74 of the pattern number 110 is extracted and passed to the generation unit 4. The generation unit 4 outputs, for example, “I feel like going to lunch.” As a result.

[0019]

As described above, according to the present invention, a clause which is not directly connected to a predicate in the source language can be analyzed by the source language
When translating a sentence that is essential in translation, it can be translated as a target language having an appropriate structure.

[Brief description of the drawings]

FIG. 1 is a block diagram of an embodiment of a natural language automatic translation apparatus according to the present invention.

FIG. 2 is a diagram showing a specific configuration example of an expression bilingual dictionary of FIG. 1;

FIG. 3 is a flowchart illustrating an operation of a conversion unit in FIG. 1;

FIG. 4 is a diagram showing an example of a tree structure as a result of source language analysis.

FIG. 5 is a diagram showing another example of a tree structure as a result of the source language analysis.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input part 2 Analysis part 3 Conversion part 4 Generation part 5 Output part 6 Analysis dictionary 7 Expression bilingual dictionary 8 Generation dictionary

────────────────────────────────────────────────── ─── Continued on the front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06F 17/27-17/28 JICST file (JOIS)

Claims (1)

(57) [Claims] A natural language for translating a source language sentence into a target language sentence
In the automatic word translator, the source language sentence to be translated is referenced by referring to the analysis dictionary.
The analysis unit analyzes the source language structure of the tree structure described in the above, and supports the expression patterns of the source language and the expression patterns of the source language.
The expression pattern of the target language to be stored is stored in the source language table.
The current pattern uses the predicate as a key word and
Has a dependency condition between components, and
Takes a tree structure of three or more layers with the predicate at the top, and
For source language constructs that contain disjoint components,
Involvement of components not directly connected to the predicate of the key word
A predicate is entered for the expression bilingual dictionary having the receiving condition and the source language structure analyzed by the analysis unit.
From the expression bilingual dictionary, the components of the source language structure
Search for source language expression patterns that satisfy elementary dependency conditions
And an object corresponding to the searched source language expression pattern.
By obtaining a language expression pattern, the original language to be translated
Convert expression patterns of word sentences into expression patterns of target language
And a conversion unit based on the expression pattern of the target language converted by the conversion unit.
The source language sentence to be translated with reference to the generated dictionary.
A natural language automatic translation device, comprising: a generation unit that generates a natural language sentence .