JPS63198162A - Context processing device for sentence generation - Google Patents

Abstract

PURPOSE:To attain the generation of a sentence which is intelligible, which has satisfactory information transmitting efficiency and which is natural including abbreviation, a pronoun and a demonstrative pronoun from a meaning expression independent of an input sentence by originally executing a pronoun processing, an abbreviation-processing and the addition-processing of demonstrative pronouns from the meaning expression to be outputted. CONSTITUTION:A declinable word case control table 2 decides the grammar relation of respective components at every even withe respect to the meaning expression consisting of plural events (a declinable word, and case components controlled by the declinable word) to be generated as an inputted sentence. With collating to a context table 3 based on the grammar relation, a subject is decided and is written in the context table 3 according to a subject decision rule stored in a subject decision rule file 4, and the decision of the subject and the pronoun processing, the abbreviation processing and the addition of the demonstrative pronouns are executed. Thus, the sentence which is intelligible, which has the satisfactory information transmitting efficiency and which is natural can be generated from the meaning expression to be generated independent of the input sentence at the time of meaning analysis.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a natural language sentence generation device, and more particularly to a sentence generation device from a semantic expression in a natural language processing device based on semantic understanding.

2. Description of the Related Art A conventional sentence generation device generates a sentence from the meaning expression of a sentence based on the order of appearance of an input sentence during semantic analysis and the sentence definition, and generates a sentence based on the meaning expression to be generated. Normally, processing such as determination, abbreviation, pronounization, and demonstrative addition is not performed, and the readability, redundancy, and consistency of the generated sentence are determined by the readability and redundancy of the input sentence. - Depends on consistency.

Problems that the invention aims to solve In conventional text generation devices, processing of the original text (summarizing, changing the text style, adding new information not found in the original text, etc.) is performed on the meaning expression based on semantic understanding. It is difficult to generate a sentence from a semantic expression that is different from the semantic expression of the original sentence, or to generate a sentence from the same semantic expression as the original sentence, with a different order of appearance from the input sentence, and a different sentence selection. .

SUMMARY OF THE INVENTION An object of the present invention is to provide a text generation context processing device that solves the above problems and enables natural text generation with high information transmission efficiency.

Means for Solving Problems The present invention determines the subject from the content (semantic expression) to be output, manages the context of the output sentence, and uses pronounization, abbreviation, and demonstrative addition based on the context. This is a context processing device for sentence generation.

Operation The present invention is based on semantic understanding (in a natural language processing device,
In order to generate natural sentences that are easy to read and have good information transmission efficiency from the semantic expressions that should be generated without depending on the input sentences during semantic analysis, the sentences that should be generated based on the semantic expressions that should be generated and the themes of the sentences that have already been generated. The subject of the sentence is determined, and the sentence to be generated is abbreviated, made into a pronoun, and a demonstrative is added based on the context information of the sentence already generated and the semantic expression of the sentence to be generated.

Example The details of the present invention will be explained below based on examples.

The figure is a configuration diagram of an embodiment of the present invention. In the figure, 1 is the context processing unit that determines the subject, abbreviates it, turns it into a pronoun, and adds demonstratives, 2 is the pragmatic case control table that determines the case of the surface sentence from the semantic expression, and 3 is the processing that has been done so far. 4 is a theme determination rule file;
5 is an abbreviation rule file, 6 is a pronounization rule file,
7 shows a demonstrative addition rule file.

The context processing unit receives a semantic expression consisting of multiple events (the predicate and the case components that are dominated by the predicate) to be generated as a single sentence by human power, determines the theme, and then converts it into pronouns, abbreviations, and demonstratives. Outputs the generation expression of the added sentence.

First, the grammatical relationship of each component is determined for each case using the pragmatic case dominance table for the input semantic expression, and based on the grammatical relationship, a context table that holds the context of sentences output so far is referred to, The theme is determined according to the theme determination rule stored in the theme determination rule file.

The theme determination file includes, for example, Rule 1: If the input meaning expression is composed of a plurality of events and the grammatical relation nominative is a common element, that is set as the theme.

Rule 2: As long as the verb expresses the existence of something, it is irrelevant to the context (the nominative is the subject).

It stores rules such as The determined theme is newly written to the context table. Further, the context table stores the themes, predicates, case elements, etc. of the sentences processed so far.

The abbreviation section refers to the context table and pragmatic case dominance table according to the abbreviation rules stored in the abbreviation rule file, and performs operations based on syntactic factors (elimination of case components, based on the previously determined theme). (abbreviation) Modifying components are abbreviated based on the context.

For example, the abbreviation rule file contains Rule A as a syntactic abbreviation rule.If the semantic expression of a sentence to be output consists of multiple events in a coordinating relationship, the case components that can be analyzed by control-adjunction are Abbreviate.

Also, based on the theme (an example of an omission rule is Rule B; if the old theme and new theme are the same, the theme component is omitted.

Also, based on the context (as an omission rule, for example, Rule C: If the context table contains content equivalent to a modifier phrase that limits or modifies a noun concept, the meaning expression corresponding to the modifier phrase is omitted.

It stores rules such as The pronominalization processing unit performs pronounization processing on the abbreviated semantic expression according to the pronominalization rules stored in the pronounization rule file.

In the pronominalization rule file, for example, Rule A: If the same element as the meaning expression of the sentence to be generated exists in the context table and can be associated with -, the element is converted into a pronoun.

It stores pronounization rules such as.

In the demonstrative addition processing unit, demonstratives are added according to the demonstrative addition rules stored in the demonstrative addition rule file, based on the process of identifying the elements of the context table and the elements of the semantic expression to be generated.

Effects of the Invention As described above, according to the present invention, without depending on the input text,
Since it is possible to independently process pronouns, abbreviations, and the addition of demonstratives from the semantic expression to be output, it is possible to create understandings including abbreviations, pronouns, and demonstratives from semantic expressions that do not depend on the input sentence, which was not possible with conventional generation devices. It becomes possible to generate natural sentences that are easy and efficient in transmitting information. Further, according to the present invention, it is possible to generate natural sentences from processed semantic expressions such as summaries that do not depend on the amount of original text, which was not possible in the past, and the effects are extremely effective in natural language processing.

[Brief explanation of the drawing]

The figure is a block diagram of a syntax processing device for text generation according to an embodiment of the present invention. ■...Context processing unit, 2...Pragmatic case dominance table, 3
...Context table, 4...Subject determination rule file,
5... File, 6... Pronounization rule file, 7
...Demonstrative addition rule file. The name of the agent is Toshio Tsuchinakao, a patent attorney.

Claims (1)

[Claims] A pragmatic case control table for determining the case of surface sentences from semantic expressions, a context table that holds context information of sentences processed so far, a theme determination rule file, an abbreviation rule file, and a pronominalization rule. file and a demonstrative addition rule file, and uses the predicate case dominance table to generate a semantic expression consisting of a plurality of events (predicates and case components dominated by predicates) to be generated as an input sentence. Determine the grammatical relationship of each component for each event, refer to the context table based on the grammatical relationship, determine a theme according to the theme determination rule stored in the theme determination rule file, and write it in the context table; A context processing device for sentence generation characterized by determining a subject, converting it into a pronoun, abbreviating it, and adding a demonstrative.