JPS63168774A - Subject supplementing system for mechanical translation system - Google Patents

Abstract

PURPOSE:To obtain a natural translation by providing a subject candidate list in order to supplement the subject of a Japanese sentence from which the subject is omitted, selecting the most suitable one therefrom and supplementing as the subject and translating. CONSTITUTION:At the time of reaching the sentence from which the subject is omitted, a subject selecting means 3 sequentially watches the subject candidate list 2 from the last part and selects one which can be the subject grammatically and in a sense. Therefore, the attribute of the reference of the selection as well as a word is stored in the subject candidate list 2. Ordinarily, since the subject of the sentence in which the subject is omitted is a lately appearing noun, when the most lately appearing noun of the nouns grammatically correctly connected nouns is used as the subject of the sentence in which the subject is not present, the correct translation can be formed. When the proper subject candidate is not found, or when there are more than two candidates and which is more suitable cannot be decided, the translation is executed as by using a passive voice.

Description

[Detailed Description of the Invention] [Summary] A method for supplementing the subject of a Japanese sentence where the subject has been omitted in a machine translation system, especially a machine translation system that performs translation from a source language such as Japanese, which often does not have a subject. , a subject candidate list is provided, from which the most probable one is selected and translated as a supplementary subject, thereby providing a more natural translation.

[Industrial Field of Application] The present invention relates to machine translation from Japanese, and relates to a method for supplementing subjects omitted in the original text.

In Japanese, the subject is often omitted. On the other hand, in Western languages, the subject is essential, so some kind of manipulation is required.

[Prior Art] Conventionally, in machine translation from Japanese, successful translation has been achieved by generating a sentence (passive) in which the object is the subject for a sentence without a subject.

FIG. 4 is a diagram showing an example of machine translation according to the prior art.

As shown in the Japanese example sentences, the first sentence a has a subject, but the following sentences c and d omit the subject.

The translation of these according to the prior art is as shown on the eye side 1. That is, the subject-less sentences S, C, and D are generated as passive sentences with the object as the subject.

In this way, machine translation using conventional methods produces sentences that are unnatural and awkward, although the translation is correct.

[Problems to be Solved by the Invention] As explained above, according to the conventional method, the translation of a sentence without a subject results in an unnatural and awkward sentence.

Therefore, it is necessary to fill in the omitted subject, but if you fill in the wrong subject, the accuracy of the translation will decrease.

The present invention aims to provide a novel subject supplementation method that solves these conventional problems.

[Means for Solving the Problems] FIG. 1 shows a principle block diagram of the subject supplementation method in the machine translation system using Japanese as the source language of the present invention.

In the figure, reference numeral 1 denotes a subject candidate list creation means, which analyzes an input sentence, extracts nouns, and stores them as subject candidates in a subject candidate list 2. The subject candidate list 2 stores newly appearing subject candidates in a connected manner.

Reference numeral 3 denotes a subject selection means, which when a sentence in which the subject is omitted is reached, selects an element of the subject candidate list 2 from the later part of the list to see if it can be the subject.

Reference numeral 4 denotes a subject complementing means, which complements the pronoun of the noun selected by the subject selecting means 3 as the subject of a sentence without a subject.

[Operation] The subject candidate list creation means analyzes the input Japanese sentence, extracts nouns, and connects them one after another in the subject candidate list as possible subjects of subsequent sentences.

When a sentence with an omitted subject is reached, the subject selection means:
Go through the list of subject candidates in order from the end and select those that can be grammatically and semantically the subject. Therefore, in addition to words, the subject candidate list includes attributes that serve as reference for selection.

Usually, the subject of a sentence with a subject omitted is a noun that has appeared recently, so if you use the noun that appears most recently among nouns that are grammatically correctly connected as the subject of a sentence without a subject, you can get a correct translation. It will be done.

When a suitable subject candidate cannot be found, or when there are two or more candidates and it is not possible to decide which one is suitable, the translation is performed using the passive voice.

Depending on the type of document, some words are particularly likely to become subjects.

For example, in a story, the main character, etc., and in an essay,
It is appropriate to give priority to related technical terms (substance names, experimental equipment, etc.).

Therefore, it is possible to cope with this by providing means for switching the selection method depending on the type of document.

[Example 1] The present invention will be explained in more detail below with reference to examples shown in FIGS. 2 and 3.

FIG. 2 is a diagram showing a description example of a subject candidate list in an embodiment of the present invention.

As shown in the figure, in addition to (1) the subject candidate word, (2) the attribute of the word, (2) the pronoun, and (3) the reference count counter are stored in this subject candidate list.

Attributes are used to check whether a sentence can be a grammatically correct subject.

When this word is selected as the subject of a sentence without a subject, a sentence is generated using this pronoun as the subject.

The reference count counter counts the number of times the subject is used as the subject of a sentence without a subject, and is used as a condition for keeping the subject candidate list when discarding the oldest one due to the capacity limit of the subject candidate list.

In the figure, the arrow → at the top indicates that new words are connected, and the arrow ← at the bottom indicates the pointer to the next word to be referenced in this list.
nil indicates that there are no more words to refer to.

The operation according to this embodiment will be described below with respect to the Japanese example sentence shown in FIG.

(1) First, when translating the first sentence a, the subject "chocho" and the object "(mandarin orange) tree 1" are added to the subject candidate list.
) has animal as an attribute and it as a pronoun, and ``tree has plant as an attribute and it as a pronoun, and ``butterfly'' is the subject. .

From then on, the reference count counter is set to 1.

(2) Next, in the translation of the sentence, it is detected that the subject of "add (something)..." is omitted. The subject selection means presumes that the subject is a volitional verb since "tsuke" is a volitional verb, and later searches the subject candidate list on the condition that it is an animal, a human, or something that is likely to be anthropomorphized. ``r 木'' is inappropriate, but r chocho J remains and there are no other words registered in the subject candidate list, so ``chocho'' is selected as the subject of the sentence. A reference count counter indicating that the word r Chocho J in the subject candidate list is used as a subject is incremented by one.

(3) The noun ``bunsu'' is extracted, and ``s to
mach) and 'leaf' are added to the subject candidate list with their respective attributes: body, plant, and the pronoun it.

(4) Next, in the translation of sentence C, “(what are you doing)?

It is detected that the subject of is omitted. The subject selection means presumes that the subject is a volitional verb since ``shiru'' is a volitional verb, and later searches the subject candidate list on the condition that it is an animal, a human, or something that is likely to be transformed into an hominin. 1
``leaf'', ``tummy j'', and ``tree'' are inappropriate, but ``chocho J'' remains and there are no other words registered in the subject candidate list, so ``chocho'' is selected as the subject of sentence C. The reference count counter for the word "chocho" in the subject candidate list is incremented by 1.

(5) Next, in the translation of sentence d, it is detected that the subject of ``j'' is omitted.
Since "I'm giving birth" is a verb with intention, it is assumed that the subject is a verb with intention, and the subject candidate list is later searched under the condition that it is an animal, a human, or something that is likely to be anthropomorphized. 1 leaf”
, ``taka'', and ``r tree'' are inappropriate, but ``chocho'' remains and there are no other words registered in the subject candidate list, so ``chocho'' is selected as the subject of sentence d. Increment the reference count counter of "rchocho" in the subject candidate list by 1.

(6) The noun r egg J (egg) in sentence d is extracted and added to the subject candidate list along with its attributes and pronouns. At this time, if the length of the candidate list is set to "4", the oldest candidate list (the word that appeared earlier)
``tree J'', which has never been referenced.
) is discarded. If there is nothing that has never been referenced, discard the oldest one.

(7) When the subject selection means cannot find a suitable candidate from the subject candidate list, or when there are two or more candidates and it cannot be determined which one is suitable, it outputs "unselectable" and translates as a passive. .

In FIG. 3, a sentence with the subject supplemented in this way and translated into English is shown as eye side 2.

It can be seen that the translation is extremely natural compared to the eye side 1 according to the prior art shown in FIG.

[Effects of the Invention] As explained above, according to the present invention, in a machine translation system from Japanese, translation accuracy can be improved by selecting the most likely omitted subject and supplementing it. , its practical effects are extremely large.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of the principle of the present invention. FIG. 2 is a diagram showing a subject candidate list according to an embodiment of the present invention. FIG. 3 is a diagram illustrating an example of translation according to an embodiment of the present invention. It is a figure which shows the translation example by a prior art. In the drawings, 1 indicates a subject candidate list creation means, 2 indicates a subject candidate list, 3 indicates a subject selection means, and 4 indicates a subject complementation means.

Claims (3)

[Claims] (1) In a machine translation system with Japanese as the source language, subject candidate list creation means (1) that analyzes an input sentence, extracts nouns, stores them as subject candidates, and manages them as a list.
, a subject candidate list (2) in which new subject candidates extracted by the subject candidate list creation means (1) are connected and stored one after another, and a subject candidate list (2) when a sentence in which the subject is omitted is reached.
), and a subject supplementary means (4) that supplements the pronoun of the noun selected by the subject selection means (3) as the subject of the subject-less sentence. What is claimed is: 1. A method for supplementing a subject in a machine translation system, characterized in that the method is configured to execute translation using a sentence supplemented with a subject. (2) The machine according to claim 1, characterized in that when the subject selection means (3) is unable to select a subject, the machine is configured to translate the object as a passive subject. Subject complementation method in translation systems. (3) The subject selection method using the subject selection means (3) above,
2. A subject complementation method in a machine translation system according to claim 1, further comprising means for switching according to the type of document.