JPH0412505B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field of the Invention The present invention relates to a machine translation device that performs foreign language translation such as from English to Japanese.

Prior Art and Problems In order to have a device perform translation, that is, conversion between different languages such as Japanese, English, and French, instead of simply obtaining the information, generally one translation device is installed, and sentences in one language are inputted. Usually, the translation device outputs a sentence in another language converted according to a predetermined conversion rule built in, and no human intervention is performed during this process. However, the same sentence can have different meanings depending on whether it is in the title, table of contents, or in the text. For example, the following sentence A, The door opened for us, can be interpreted as the door opened for us, or the door opened for us.
If opened is a transitive verb, then it is an intransitive verb. From this sentence alone, it is not possible to distinguish between an intransitive verb and a transitive verb, and it is impossible to determine whether it is or. However, if sentence A is in a title, the title is usually a noun, and if it is in a sentence, the probability is high because the sentence is usually made up of a subject-predicate relationship.

The above-mentioned problem of ambiguity regarding sentence A arises from the multi-part nature of English words, but of course ambiguity arises for various reasons. For example, it is common for there to be multiple Japanese nouns for English nouns, and depending on how you choose them, there can be two or three different interpretations.

Purpose of the Invention The present invention attempts to deal with the ambiguity problem in machine translation, and attempts to make accurate translation possible by extracting sentence types from format information and effectively utilizing this information. It is.

Structure of the Invention The present invention is a machine translation device that performs foreign language translation such as from English to Japanese, which divides input sentence data to be translated into sentences one by one, and determines whether the sentence is a title or bulleted list based on the format of each sentence. , a sentence type detection means for determining a sentence type such as whether it is a main text, a plurality of translation mechanisms provided corresponding to the sentence types output by the sentence type detection means and performing translation processing according to the sentence type; Translation mechanism selection means for inputting the sentence data to a translation mechanism suitable for the sentence data according to the sentence type signal of each sentence outputted by the sentence type detection means. Explain.

Embodiment of the Invention FIG. 1 shows an outline of a machine translation device according to the present invention. As shown in the figure, this device includes a sentence type detection unit 10 that separates input sentences into sentences and determines the sentence type; In this example, there are three types of translation mechanisms 14a, 14b, 14c, and the detection unit 1.
Translation mechanism 1 to be used according to the sentence type detected by 0
It consists of a translation mechanism selection section 12 that selects 4a to 14c. A major feature of the present invention is that not one type but multiple types of translation mechanisms are provided. In this example, the plurality of types of translation mechanisms 14a to 14c are 14a for titles, 14b for main texts, and 14c for command sentences, but these are preferably classified and selected according to the characteristics of the language to be translated. In addition, imperative sentence translation mechanism 1
Regarding 4c, there is a risk of mistranslation of an imperative sentence unless one is aware that it is an imperative sentence. For example, SET DIAL would be a command sentence to set the dial, and a plain sentence would be a dial for setting. This type of command statement appears frequently in machine maintenance manuals and the like. Therefore, if it is expected that the language to be translated includes imperative sentences, it is advisable to provide a translation mechanism 14c so that one can be selected. Of course, if there is no risk that this kind of thing will be included in the target language, translation mechanism 1
There is no need to prepare 4c.

A translation mechanism is a device whose purpose is to convert a text written in a source language into a specific other language (target language), and is generally equipped with a dictionary and conversion rules. is divided into words, each word is associated with grammatical attributes and translations using a dictionary, processed such as changing the word order using conversion rules, and then output in the target language. There are many known translation mechanisms themselves.
Adopting an appropriate one of these translation mechanisms,
Translation mechanisms 14a to 14c with modifications as necessary
And it is sufficient.

In this device, the input sentence, that is, the language to be translated IS, is input to the sentence type detection unit 10, the input sentence is cut out sentence by sentence, and the type of sentence that each sentence is is determined.
It detects whether it is a title, a figure explanatory text, or a part of a table of contents, and sends that one sentence S to the selection section 12 together with a sentence type signal KS. The selection unit 12 selects one of the translation mechanisms 14a to 14c according to the sentence type signal KS, and sends the sentence S sent from the detection unit 10 to the selected translation mechanism that is most suitable for the translation. For example, if the sentence S is the title, the sentence is translated by the translation mechanism 14a
If it is a main text, it is sent to the translation mechanism 14b. TS indicates the output sentence translated in this way, that is, the translated sentence.

The sentence type detection unit 10 separates input sentences every minute,
The format of each sentence is analyzed to determine the sentence type. Here, the input sentence is assumed to be English, the output sentence is Japanese, and the format is whether there is a column indentation to indicate the beginning of the sentence, whether it ends with a period, or whether there are continuous blank lines. , whether each word starts with a capital letter, etc. Text types include titles, bulleted lists, tables of contents, main text, figures, tables, etc., but classifications such as main text and other text, external text, imperative text, and other types are also valid. The translation mechanism is configured by sentence type, but 1
There is no need to limit the type of sentence; for example, titles and tables of contents are usually converted into nouns (noun phrases), so they can be handled by the same translation mechanism. Next, the procedure for extracting each sentence from the input sentence and determining the sentence type of the extracted sentence will be explained with reference to FIG.

In FIG. 2, reference numeral 20 denotes a document to be translated, and sentences written in this document are read photoelectrically and character-recognized to become a sentence to be translated, that is, the input sentence IS described above. The method of inputting the sentence to be translated is not such a photoelectric method, but may also be a method of reading what is recorded on a magnetic tape, punch card, or the like. However, since the present invention is based on translating what is written in a document, it is necessary to be able to identify each line regardless of the input means. In the present invention, the next line is always referred to when extracting a sentence. 22 is the current row buffer, 24
are next line buffers, into which character data of one line of the input sentence and the next line are input. Sentence extraction and sentence type discrimination are performed from the character data of these buffers, and these are mutually related. For example, sentences usually end with a period, but some sentences end with an exclamation point. Therefore, this device uses the following criteria for sentence extraction.

(1) There are two or more spaces after a period, exclamation mark, or question mark.

(2) The next line is in a specific format.

The specific format here refers to the following: Blank lines, headings, bulleted lists, headings with section numbers,
Table of contents, paragraphs.

Of course, a period does not mean the end of a sentence; for example, when multiple nouns are listed by their first letter, they are used as SH, so it is not reliable to detect the end of a sentence with just a period. By the way, there is a convention to leave two characters blank after a period, exclamation mark, or question mark at the end of an English sentence, so the above
Using criterion (1), the end of a sentence can be detected. The period attached to the first letter above does not leave a space after it, or only leaves a space between one character and not two characters. Also, some sentences do not have a period at the end. For example, some bulleted lists are (1)...(2)... without a period at the end of the sentence, and if a period is used as a condition for detecting the end of a sentence, all the sentences in the bulleted list will be concatenated. Also, some sentences that lead to bullet points (sentences that precede bullet points) do not have a period, and such sentences will also be connected to bullet points if a period is the end-of-sentence detection condition. By setting (2) above as the sentence extraction criterion, such a problem can be solved.
A "blank line" in a specific format is a line where nothing is written,
This is an empty line, and the current line can be considered to end when the next line is empty. Furthermore, since the title itself is an independent sentence, even if the current line lacks the condition (1) above, if the next line is the title, the current line can be considered as the end. This applies hereafter.

To explain how to detect a specific format,
A blank line can be detected by the fact that there is no character data on the line, and a heading can be detected by the fact that the sentence ends without the condition (1) above, and the next line is in a specific format such as a blank line or bulleted list. It can be detected by the fact that it exists.
A bulleted list can be detected by a number followed by a period or a number in parentheses, and a section heading with a number can be detected by a number or a repeated number and period (1 or 1, 1, 2, etc.). It can be detected by the following sentences. A table of contents can be detected by the title or section numbered title, symbol repetition, and numbers, and a paragraph can be detected by the fact that the starting line is to the right of the previous line (the beginning of the calligraphy is downward).

Depending on the document to be translated, some of the specific formats may be meaningless, such as the document does not have a table of contents. The format selection table 26 deals with this, and specifies which of the above-mentioned formats to be used by the sentence type determining means 28 for each translation target, and which formats are valid and which are invalid. This specification is made by the operator. The sentence type discrimination means 28 and the sentence extraction means 30 use the character data of the current line buffer 22 and the next line buffer 24 to discriminate the sentence type and extract the sentence as described above, and the extracted one sentence is passed through the sentence buffer 32 to As the input sentence S, the sentence type is sent to the above-mentioned sentence type signal through the sentence type buffer 34.
It is input to the translation mechanism selection unit 12 as KS.

Effects of the Invention As explained above, in the present invention, the translation mechanism corresponds to (adapts) text types such as body text and title.
Multiple types are provided, and the target sentence for translation is a sentence without a period,
The type of sentence is determined from the format, such as, and inputted to the corresponding translation mechanism, where it is translated, so even sentences that can be interpreted in two or three ways can be translated into the appropriate translation, and the machine translation device There are some great things that contribute to improved performance.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, and FIG. 2 is a block diagram showing details of a part of FIG. In the drawing, 10 is a sentence type detection means, 14a to 14c.
12 is a translation mechanism selection means;
IS is input sentence data, S is sentence data, and KS is a sentence type signal.

Claims (1)

[Claims] 1. In a machine translation device that performs foreign language translation such as from English to Japanese, input sentence data to be translated is divided into sentences,
A sentence type detection means is provided to determine whether the sentence is a title, bulleted list, main body, etc. from the format of each sentence, and a sentence type is provided corresponding to the sentence type output by the sentence type detection means. a plurality of translation mechanisms that perform translation processing according to the sentence type detection means, and a translation mechanism selection means that inputs the sentence data to a translation mechanism suitable for the sentence according to the sentence type signal of each sentence outputted by the sentence type detection means. A machine translation device characterized by: