JPH01209571A - Generating method for japanese word - Google Patents

Abstract

PURPOSE:To obtain the sentences of proper Japanese words by defining the composition of a numeral and an auxiliary numeral as translated words in case a noun has an auxiliary numeral and putting a numeral into a position right before the noun in case this noun has no auxiliary numeral for execution of translation. CONSTITUTION:In case an element in a sentence structure includes a noun and a numeral which modifies said noun, the processing method is changed according to the presence or absence of the auxiliary numeral information. In other words, the composition of a numeral and an auxiliary numeral as the translated words in case the noun has no numeral. While a numeral is put into a position right before a noun for translation in case the noun has no auxiliary numeral. As a result, a noun can be translated in the natural form of a Japanese word in any case and the quality of the translated sentences, etc., can be improved.

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field The present invention relates to a Japanese language generation method in an English-Japanese machine translation system, an automatic response conversation device, etc.

Conventional technology In general, English-Japanese machine translation systems, etc., successfully generate Japanese translations for nouns with particles, such as "two beautiful dogs," but for nouns without particles, they produce Japanese translations such as "two beautiful buildings."'' or ``No. 2 Beautiful Building'', which is a literal Japanese translation, which is very unnatural. In other words, in these cases, it is natural to translate it as ``two beautiful buildings.''

Therefore, we forced the number particle "r" to accompany [two ~
There is also a way to translate it like this. However, in the case of a number word of 10 or more, the translation would be something like "20 beautiful buildings".
This is unnatural in Japanese.

That is, when translating a noun accompanied by a number word, it is necessary to change the notation to be generated and translated depending on whether or not the noun is accompanied by a number word.

Purpose The present invention was made in view of the above points, and when translating a noun accompanied by a number word, it is possible to create a proper Japanese sentence that does not become unnatural as Japanese depending on whether or not the noun is accompanied by a particle. The purpose is to provide a Japanese generation method that can generate Japanese.

Structure In order to achieve the above object, the present invention has relationship information between elements in a sentence structure, notation of each element, part of speech, and conjugation information,
Furthermore, if the element is a noun, it has its particle information,
In a Japanese generation method in a system equipped with a means for recognizing Japanese structures, when an element in a sentence structure includes a noun and a number word that modifies this noun, if the noun has a particle, the number word and particle are combined. This method is characterized in that the number word is translated by combining the number words, and when the noun does not have a particle, the number word is inserted immediately before the noun.

Hereinafter, one embodiment of the present invention will be described with reference to the drawings. This embodiment is a Japanese language generation method applied to, for example, an English-Japanese machine translation system, and an overview of the system is shown in the block diagram of FIG. First, an English sentence to be translated is input into an analysis device 2 from an input device 1 such as a keyboard. This analysis device 2 performs morphological and syntactic analysis of English sentences by referring to the search results of a dictionary search device 4 that searches a dictionary device 3 equipped with a word dictionary, an analysis grammar, and the like. Then, the result is syntax-converted by the conversion device 5 into a form that is easy to translate. That is, it is output to the generation device 6 as a Japanese basic structure tree as shown in FIGS. 3 and 5. - The processing in this generation device 6 is the Japanese language generation method that characterizes this embodiment. Here, each node in the Japanese basic structure tree has already been given dictionary information from the dictionary device 3 as shown in FIGS. 4 and 6 by the conversion device 5. The output from such a generating device 6 is output to an output device 7 such as a CRT. Further, each device is controlled by a control device 8.

The processing method in the generation device 6 will now be explained. First, this generation device 6 sequentially traces the nodes from the root node and uses them according to the structure.

Adding auxiliary verbs, particles, etc., replacing tree structures, etc. (
This is the first phase, the outline of which is shown in the flowchart of FIG. 7). After that, by sequentially tracing the nodes and outputting the notation of its own child nodes, and after that output, recursively outputting the notation of its own node (tracing the nodes top-down, left-to-right) (the second phase), and finally the Japanese translation is generated (second phase).

In the first phase of sequentially tracing nodes in this way, if the node is other than a noun node, a predicate node or other node is activated and processed as shown in FIG. Here, processing of predicate nodes means processing of nodes such as verbs and adjectives that are predicates in Japanese.

Further, processing of other nodes means that, for example, changing the position of a predicate node of a child node is performed depending on the type of node.

On the other hand, in the case of a noun node, characteristic noun node processing of this embodiment including processing of particles is activated. That is, the combination of "noun node processing" and "counter word processing" shown in FIG. 8 means noun node processing, and other processing may exist after the particle processing.

Therefore, the particle processing activated as part of the noun node processing will be explained with reference to the flowchart of FIG. First, if there is no numeral node among the child nodes of the currently focused noun node, the second process ends without performing anything.

Consider a case where the noun node of interest has a numeral node as a child node. First, when the number node itself has particle information, for example, the particle information "one person" shown in Figure 4, no special particle processing is required. It is a composite, that is, ``number + particle particle''. As a result, a translated word such as "two boys" is generated, for example.

On the other hand, when a number node has a number node but does not have particle information, for example, when the content is in the dictionary device 3 as shown in Figure 6, not only the generated notation is changed, but also the word order of the number word is changed. . That is, the notation of a number node and the notation of a noun node are used to generate a notation of a noun node in the form of "number ten noun". Specifically, we absorb numbers as information from number nodes, and for noun nodes, we change the translation notation of our own nodes to ``The absorbed number 10 yuan self (
``translation of the noun)'' and delete the numeral node from the sentence structure. In other words, the number held by the numeral node is inserted immediately before the noun. As a result, "2 boys"
A translation like this is generated.

After the first phase processing is completed, top-down 1eft
The node is traced to right, and a Japanese sentence is finally generated.

Now, in Figure 3, when there is a number node with particle information as shown in Figure 4, the particle information ``1 person'' is added by the particle processing in Figure 1 and the word order is translated without changing. The Japanese basic structure tree of is shown.

On the other hand, FIG. 5 shows a Japanese basic structure tree in the case where, as shown in FIG. 6, there is no particle information, the word order is changed by the particle processing shown in FIG. 1 and the word is translated without particle information.

As can be seen from these examples, even if the sentences are the same in English, depending on whether or not the number words have particle information, they can be interpreted as ``two handsome American boys'' or ``a handsome American boy''. It is translated as natural Japanese, as in ``2 Shonen''.

In this embodiment, a tree structure is used as the structure of the generation method, but the present invention is not limited to this, and can easily be applied to a network-like structure, for example. Furthermore, it is also possible to have particle information and the like not as node information but as information in the dictionary device 3, and to search and use this. Furthermore, the tree structure input to the generation device 6 is not limited to the one in the translator, but may be, for example, the logical structure of the answer of the makeshift system.

Effects As described above, when an element in a sentence structure includes a noun and a numeral that modifies this noun, processing is changed depending on whether or not the noun has particle information, and when the noun has particle information, the numeral is The translation of the number word is made by combining it with a particle and a particle, but if the noun does not have a particle, the number word is inserted immediately before the noun to create a translation that is not unnatural as Japanese in any case. The noun method can be generated using , and the quality of translated sentences can be improved.

[Brief explanation of the drawing]

The drawings show one embodiment of the present invention, FIG. 1 is a flowchart showing an outline of particle processing, FIG. 2 is a block diagram of the entire system, and FIG. 3 is a tree structure diagram showing a specific example when particle information is included. , Figure 4 is an explanatory diagram of the contents of the corresponding translation dictionary,
Figure 5 is a tree structure diagram showing a specific example when no particle information is included, Figure 6 is an explanatory diagram of the corresponding translation dictionary contents, Figure 7 is a flowchart of the first phase, and Figure 8 is a flowchart of node processing. It is. Applicant Ricota Co., Ltd.

Claims (1)

[Claims] Japan in a system that has relationship information between elements in sentence structure, notation of each element, part of speech, and conjugation information, and furthermore, if the element is a noun, has particle information for that element, and has a means to recognize Japanese structure. In the word generation method, when an element in a sentence structure includes a noun and a number word that modifies this noun, and the noun has a particle, the combination of the number word and the particle is the translation of the number word, and the noun is a particle. A Japanese language generation method characterized in that when the noun does not have a numeral, the numeral is inserted in the position immediately before the noun and translated.