JPH0412871B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a two-word relationship extraction method for extracting dependent two-word relationships between nouns and verbs from a large amount of natural language sentences.

Practical translation methods used in machine translation, for example, Japanese-English machine translation, can be roughly divided into two methods: the universal meaning expression method (pivot method) and the interlingual correspondence conversion method (transfer method).
In any of these methods, it is necessary to use the relationship between two words in a natural language sentence in the machine translation. Therefore, in a machine translation device, the relationship between two words in a natural language sentence must be registered in a two-word relationship database in advance.

[Conventional technology]

While such two-word relationship databases do not exist in the world like Japanese dictionaries or thesauruses, the amount of data that must be registered in them is large. Traditionally, creating this type of database has relied on humans or on devices that examine collocations in natural language sentences.

[Problem that the invention seeks to solve]

It has been almost impossible to obtain the required large amount of data using the technique described above that relies on the former for registering the relationships between two words in a natural language sentence in a database.

Furthermore, with the latter technique, it is possible to examine the noun-sa verb relationship in Japanese sentences, which is the most important noun-verb relationship that is extracted as a two-word relationship. -There is a problem that the relationship between main verbs cannot be investigated, and furthermore, the relationship involving particles cannot be investigated.

[Means for solving problems]

The present invention provides a two-word relationship extraction method capable of solving the above-mentioned problems, and the method includes a natural language database that stores natural language sentences, and a natural language sentence read from the database that is converted into words. a word division means for dividing a sentence into words; a two-word relationship extraction means for performing clause synthesis on a sentence divided into words; and a two-word relation extraction means for extracting a dependent relationship between a noun and a verb from the clause; and a two-word relationship database that stores two-word relationship data indicating a relationship that can be modified between words.

[Effect]

According to the method of the present invention, a natural language sentence is divided into words, clauses are synthesized for the sentences divided into words, and dependable relationships between nouns and verbs are extracted from the clauses. Obtain the most important noun-verb dependency relationships from a large number of natural language sentences, and use this information to objectively judge the noun-verb dependency relationships. can be provided.

〔Example〕

Embodiments of the present invention will be described below with reference to FIGS. 1 to 3.

FIG. 1 shows an embodiment of the invention. In this figure, 1 is a natural language sentence database (Japanese sentences will be explained below), and 2 is a word dividing device that divides Japanese sentences into words while referring to a Japanese dictionary 3 (this device is already in production). proposed by the applicant). Reference numeral 4 denotes a two-word relationship extraction device, which performs phrase synthesis on sentences divided into words, and extracts dependable relationships between nouns and verbs from the phrases. 5
is a two-word relationship database that stores two-word relationship data indicating a dependent relationship between extracted nouns and verbs. Reference numeral 6 denotes a search device that performs search processing on the two-word relationship database.

FIG. 2 is a detailed configuration diagram of the two-word relationship extraction device 4. As shown in FIG. A phrase synthesis device 10, an analysis device 11,
It consists of an extraction device 12. The phrase synthesis device 10 is
Clause synthesis is performed by connecting adjuncts in sentences divided into words by the word dividing device 2 to independent words on the left (or connecting prefixes to independent words on the right). The analysis device 11 applies a context-free grammar (known) to the synthesized phrase, verb phrase → verb...verb phrase → conjunctive clause, and verb phrase... and creates an analysis tree until the application is no longer applied. Perform the desired process. The extraction device 12 extracts and outputs two-word relationship data indicating a dependent relationship between a noun and a verb from the obtained analytic tree.

Next, the process of extracting the relationship between two words under the above configuration will be explained.

For convenience of explanation, the following Japanese sentence will be used as an example.

(Example) This LSI can achieve more than 10 times the packaging density, contributing to the miniaturization of devices.

This example sentence is read from the Japanese sentence database 1, and is word-divided by the word division device 2 as follows while referring to the Japanese dictionary 3.

This |LSI| can |achieve|achieve |10|times|more than|times the |packaging|packaging|density|and contributes to |the miniaturization|of |devices|.

The sentence divided into words is given to the two-word relationship extraction device 4, and the clause synthesis device 10 connects the adjunct to the independent word on the left (the prefix is connected to the independent word on the right) to create a clause. Perform composition to obtain the following clause.

This |LSI| can achieve |a mounting density |of more than 10 times |and contributes to |the miniaturization of |devices.
| This phrase-synthesized sentence is input to the analysis device 11. | The analysis performed by the analysis device 11 is to apply the above-mentioned context-free grammar to the phrase-synthesized sentence from right to left of the sentence. When these context-free grammars cannot be applied, the analysis device 1
The analysis process using 0 ends. For example, in the clause-synthesized sentence above, as shown in Figure 3,
When context-free grammar is applied to the verb ``contribute.'', this is treated as a verb phrase. When the context-free grammar is applied to this verb phrase and the conjunctive phrase "to miniaturize," both are treated as verb phrases and a parse tree is generated. The verb phrase in this latest generated parse tree and the next clause in the clause-combined sentence, that is, the adjunctive clause, cannot be analyzed using the context-free grammar even if the context-free grammar is applied. The analysis process by the device ends.

An analysis tree obtained through such analysis processing is constructed, and two-word relationship data indicating a dependent relationship between a noun and a verb is left. In the example sentence above, it contributes to miniaturization.

leave. In this two-word relationship data, "miniaturization" is a noun, and "contributing" is a verb.

Each data element of the two-word relationship data thus obtained is readably stored in the two-word relationship database 5 in a predetermined association. The above data elements are stored as follows.

(Naturalization, contribution, etc.) In the above example, the natural language sentence was explained as a Japanese sentence, but the natural language sentence may be in another language, and in that case, the combination of a noun and a verb is also possible. The relationship between two words is required.

〔Effect of the invention〕

As explained above, according to the present invention, it is possible to process and acquire the most important two-word relationship, the relationship between a noun and a verb, which can be dependent on a large number of natural language sentences. Effects such as being able to objectively set the case of familiar verbs can be obtained.

[Brief explanation of drawings]

FIG. 1 is a diagram showing an embodiment of the present invention, FIG. 2 is a detailed diagram of a two-word relationship extraction device, and FIG. 3 is a diagram showing an example of an analytic tree. In the figure, 1 is a natural language sentence database, 2
3 is a word segmentation device, 3 is a Japanese sentence dictionary, 4 is a two-word relationship extraction device, and 5 is a two-word relationship database.

Claims (1)

[Claims] 1. A natural language database that stores natural language sentences; a word dividing means that divides the natural language sentences read from the database into words; and a phrase synthesis unit that performs phrase synthesis on the sentences that have been divided into words, and extracts nouns and verbs from the phrases. and a two-word relationship database that stores two-word relationship data indicating a dependent relationship between the extracted noun and the verb. A method for extracting relationships between two words, which is characterized by being constructed in advance.