JPH02103662A - Sentence dividing system - Google Patents

Abstract

PURPOSE:To improve a sentence structure analyzing speed, efficiency, and accuracy of the title system without spoiling the information held by an original text by mechanically dividing the original text to be processed. CONSTITUTION:A translating device having a dictionary consulting system is provided with a CRT 1, keyboard 2, OCR 3, input document 4, spell checking section 5, preediting section 6, translating main body section 7, post-editing section 8, dictionary 9, grammatical rule 10, output document 11, and printer 12. When three marks and words respectively held in a punctuation mark table, coordinate conjunction table, and personal pronoun table continuously appear in an input text, it is estimated that punctuation exists between the punctuation mark and personal pronoun and the first letter of the coordinate conjunction is capitalized by considering that the pronoun is the leading word of the punctuated sentence after rewriting the punctuation mark to a sentence ending mark. Since a given text is divided in accordance with the morphological feature of the sentence in such way, accurate, quick, and efficient analyses can be performed.

Description

[Detailed Description of the Invention] The present invention provides a sentence segmentation method, more specifically, a language understanding device,
The present invention relates to a morphological analysis unit and a syntactic analysis unit in a machine translation device.

1. In recent years, natural language analysis, which is used in machine translation systems, has made remarkable progress. However, the situation is still not satisfactory in terms of accuracy, speed, etc. One reason for this is that the sentence to be analyzed is too long.

Generally, the unit of syntax analysis is one sentence. When performing analysis, it is desirable that the analysis range be short. In other words, if a shorter range is to be analyzed, the number of applications and combinations of analysis rules can be reduced, making the analysis easier.
Ambiguity generated during the analysis process can be minimized.

However, actual texts do not only consist of short sentences; on the contrary, many of them consist of long sentences. So, if you divide one sentence into multiple sentences,
Each sentence becomes a short sentence, the parsing range is limited, and the number of application and combinations of parsing rules is significantly reduced, which reduces unnecessary rule applications and combinations, improves parsing efficiency and speed, and reduces ambiguity. This also improves analysis accuracy.

As such a sentence division method, Japanese Patent Application Laid-Open No. 62-16317
In Publication No. 6, randJrorJrh
Sentence editing devices, such as overheaderJ, have been proposed that extract predetermined character string patterns and classify original sentences by having an operator provide usage suggestions. Further, Japanese Patent Application Laid-Open No. 10267/1983 discloses a machine translation device that divides an original text at locations specified by an operator's instructions, and that concatenates multiple sentences into a single sentence. .

However, these conventional devices all rely on instructions from an operator and do not perform automatic processing, so they have drawbacks in terms of appropriateness and speed of instructions. As mentioned above, when analyzing a long text, instructions from the operator are required to divide the text into segments, and the current situation is that the accuracy of analysis by machines alone without human intervention is still low.

The present invention was made in view of the above-mentioned circumstances.
The purpose of this method is to enable automatic segmentation by dividing a given text based on its morphological features, and to enable more accurate, faster, and more efficient analysis.

composition.

In order to achieve the above object, the present invention provides a delimiter table and a delimiter table having delimiters such as commas and dashes to separate input linguistic text in a morphological analysis unit in a natural language analysis system such as machine translation. It has a coordinating conjunction table with conjunctions and a personal pronoun table with nominative personal pronouns, and when the words of each table appear one after another in the input text, the delimiters and coordinating conjunctions are displayed. It is characterized by estimating that there is a break in the content between words, replacing the delimiter with a sentence-final symbol, and capitalizing the first letter, assuming that the first-coordinate conjunction is the beginning of a separated sentence. . Hereinafter, the present invention will be explained based on examples.

FIG. 1 is a block diagram showing an embodiment of a translation device equipped with a dictionary lookup method according to the present invention. In the figure, 1 is a CRT, 2 is a keyboard, 3 is an input document 0CR14, 5 is a spell check unit, 6 is a pre-editing section, 7 is a translation main section, 8 is a post-editing section, 9 is a dictionary, 10 is a grammar rule, 11 is an output document, 12 is a printer, and the information obtained by file input, keyboard input, or OCR input is The input sentences are preprocessed using spell checking and pre-editing, the output sentences obtained by the translation section are edited using the translation information in post-editing, and the input and output sentences are printed using a printer. .

FIG. 2 is a diagram showing the flow of the translation main body section. This translation main body section (translation section) 7 is roughly divided into morphological analysis, syntactic analysis,
It consists of four processes: conversion and generation. First, the morphological analysis section performs a dictionary lookup of the input text to obtain information on six individual words, and the syntactic analysis section performs parsing according to grammatical rules. Create a tree structure from the analysis results. The conversion unit transforms the tree structure of the input language into the book structure of one output word. The generation section translates the obtained tree structure node by node.

The present invention belongs to the above-mentioned morphological analysis section.

Here, the input text is English. The morphological analysis section performs the division process shown in FIG. 3 on the input text.

Figures 4 to 6 are examples of tables used in the division process in Figure 3. Figure 4 is a delimiter table, Figure 5 is a coordinate conjunction table, and Figure 6 is a nominative personal pronoun. It's a table.

Figure 3 is a diagram showing the flow of the division process, in which the input text is examined from beginning to end in increments of l ii.If it is not a 0 delimiter, if it is not a coordinating conjunction, or if it is not a nominative personal pronoun, Move the pointer forward one word to find out. If it is a delimiter, advance the pointer one word, if the next word is a coordinating conjunction, advance it one more word, and if the next word is a nominative personal pronoun, that is, a delimiter, a coordinating conjunction, and a nominative personal pronoun. It is determined that there is a break in the text when three curses occur in succession. At this time, the text is divided between the delimiter and the coordinating conjunction. At the same time, the delimiter is the end of sentence symbol. If the text is in English, use a period. Coordinating conjunctions begin the next sentence, so capitalize the first letter.

This will be explained below using an example.

Assume that the following text (A) is now given.

(A), The arrival of export
ts from the Middle East
I'll be a 5hock, and I don't
t see any organized way to
absorb it.

The checking procedure is performed according to the following (1) to (8).

(1) Check each word from the beginning of the text to see if it is a delimiter.

Starting from rTheJ and ending with rshockJ, there is no delimiter, so this process is repeated word by word.

If it is a delimiter, look at the next word.

The “,” next to rshockJ is a delimiter, so
Move on to (3).

(2).

Find out whether the following words are coordinating conjunctions.

Check randJ.

If it is a coordinating conjunction, look at the next word6 Since randJ is a coordinating conjunction, Move on to (5).

Check whether the following word is a nominative personal pronoun.

Look up the next word "I".

(6) If it is a nominative personal pronoun, a division process is performed.

Because “tech” is a nominative personal pronoun, Move on to (7).

Change the delimiter to the end of sentence symbol. In this case, use a period.

Change "," to ",". In other words, "," and the next word ra
Divide the sentence between ndJ and ndJ.

Capitalize the first letter of a coordinating conjunction.

As a result of the above division, randJ becomes the beginning of a sentence, so the first character "a" is changed to a capital letter and becomes rAndJ.

(3).

(4).

(5).

(7).

(8).

As a result of the above division processing, 5 sentences (A) are converted into sentences (B) as follows.
) and sentence (C).

(B), The arrival of export
ts from the MIddleEast wi
ll be a 5hock.

(C), And I don't see any
organized way.

absorb it.

Note that since this method uses only morphological features,
The nominative personal pronoun table has only words that are used only in the nominative. Therefore, in the case of English sentences, for example, you, which indicates the second person, is not included in this table because it is difficult to tell whether it is the nominative case or the objective case based on its appearance (form).

As is clear from the above explanation, according to the present invention, by mechanically dividing the text to be processed, the speed and efficiency of parsing can be improved without losing the information of the original text. and accuracy can be improved.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of a translation device equipped with a dictionary lookup method according to the present invention, FIG. 2 is a diagram showing the flow of the translation main body, and FIG. 3 is a flow diagram of the 1-division processing. 4 to 6 are diagrams showing examples of tables. 1...CRT, 2...Keyboard, 3...OCR
, 4... Input document, 5... Spell check section, 6.
...Previous editorial department. 7... Translation body part, 8... Post-editing part, 9... Dictionary, 10... Grammar rules, 11... Output document, 12...
・Printer. Figure Figure Figure Figure

Claims (1)

[Claims] 1. In the morphological analysis section of a natural language analysis system such as machine translation, in order to separate input language text, a delimiter table with delimiters such as commas and dashes, a coordinate conjunction table with coordinating conjunctions, and personal pronouns are used. personal pronoun tables with the nominative case of A sentence division method that is characterized by estimating the delimiter, rewriting the delimiter with a sentence-final symbol, and capitalizing the first character of the coordinating conjunction, assuming that it is the beginning of the delimited sentence.