Publication: Generación de recursos lingüísticos mediante la extracción de relaciones entre conceptos
Loading...
Official URL
Full text at PDC
Publication Date
2017
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
Hoy en día disponemos de una gran cantidad de información en Internet y sería muy positivo poder obtener beneficio de ella para desarrollar sistemas de generación de texto. Nuestro objetivo inicial es desarrollar un generador de recursos lingüísticos basado en asociaciones de conceptos. Un generador de lenguaje natural normalmente tiene acceso a un gran conjunto de conocimiento del que selecciona información y presenta en diferentes formas a los usuarios finales. Por lo tanto, un sistema generador de lenguaje natural debe escoger información de la base de conocimiento, decidir cómo organizarla, y determinar cómo producir el texto en lenguaje natural.
El problema ante el que nos encontramos en muchas ocasiones es la desorganización de tanta información disponible en Internet, pues para que resulte útil en el desarrollo de sistemas de Generación de Lenguaje Natural, la información debe estar organizada y relacionada. Esto se está cambiando con la llegada de herramientas relacionadas con la Web Semántica. En este trabajo se va a implementar un generador que, a partir de la información obtenida de algunas de estas herramientas, aquellas relacionadas con la asociación de conceptos semánticos, desarrolle recursos lingüísticos que serán posteriormente evaluados. El objetivo final de este trabajo es conocer el valor de las asociaciones de palabras para la generación de diferentes recursos de lenguaje natural.
En un primer momento se ha desarrollado un generador de adivinanzas que crea acertijos sobre conceptos representados como sustantivos comunes. Estos acertijos se basan en comparaciones entre el concepto destino (por ejemplo, un diamante) y otras entidades que comparten algunas de sus propiedades (por ejemplo, el cemento). Los acertijos resultantes se componen como una secuencia de comparaciones utilizando esta plantilla: “Qué es … tan atributo como concepto?”, donde “atributo” es una propiedad del concepto buscado que es la respuesta al acertijo, y “concepto” es una entidad diferente que comparte el valor del atributo con el concepto destino. Por ejemplo, “¿Qué es … tan duro como el cemento y tan transparente como el pelo?” es un acertijo generado para el concepto diamante por el generador de adivinanzas.
Posteriormente, se analiza la relación entre asociaciones de palabras ya disponibles en diferentes recursos web y el tipo de guras retóricas que las personas emplean regularmente. El objetivo es identificar regularidades o heurísticas que pueden ser empleadas en una etapa posterior para desarrollar un generador de texto. Será estudiada la capacidad de este sistema para la generación automática de analogías, símiles y metáforas.
Nowadays we have a lot of information on the Internet and it would be very positive to be able to benefit from it to develop systems of text generation. Our initial goal is to develop a language resource generator based on concept associations. A natural language generator usually has access to a large set of knowledge from which it selects information and presents it to end users in different forms. Therefore, a natural language generator system must choose information from the knowledge base, decide how to organize it, and determine how to produce text in natural language. The problem we are faced with on many occasions is the disorganization of so much information available on the Internet, because to be useful in the development of Natural Language Generation systems, the information must be organized and related. This is changing with the arrival of tools related to the Semantic Web. In this work we will implement a generator that, based on the information obtained from some of these tools, those related to the association of semantic concepts, develop linguistic resources that will be further evaluated. The final aim of this work is to know the value of word associations for the generation of different natural language resources. Initially a riddle generator has been developed that creates riddles about concepts represented as common nouns. These riddles are based on comparisons between the target concept (i.e. a diamond) and other entities that share some of their properties (i.e. concrete). The resulting riddles are composed as a sequence of comparisons using this template: “What is … as attribute as concept?”, where “attribute” is a property of the searched concept that is the answer to the riddle, and “concept” is a different entity that shares the value of the attribute with the target concept. For example, “What is … as hard as concrete and as transparent as hair?” is a riddle generated for the concept diamond by the riddle generator. Later, the relation between words associations already available in different web resources and the type of rhetorical figures that people use regularly is analysed. The aim is to identify regularities or heuristics that can be used in a posterior stage to develop a text generator. The capacity of this system will be studied for the automatic generation of analogies, similes and metaphors.
Nowadays we have a lot of information on the Internet and it would be very positive to be able to benefit from it to develop systems of text generation. Our initial goal is to develop a language resource generator based on concept associations. A natural language generator usually has access to a large set of knowledge from which it selects information and presents it to end users in different forms. Therefore, a natural language generator system must choose information from the knowledge base, decide how to organize it, and determine how to produce text in natural language. The problem we are faced with on many occasions is the disorganization of so much information available on the Internet, because to be useful in the development of Natural Language Generation systems, the information must be organized and related. This is changing with the arrival of tools related to the Semantic Web. In this work we will implement a generator that, based on the information obtained from some of these tools, those related to the association of semantic concepts, develop linguistic resources that will be further evaluated. The final aim of this work is to know the value of word associations for the generation of different natural language resources. Initially a riddle generator has been developed that creates riddles about concepts represented as common nouns. These riddles are based on comparisons between the target concept (i.e. a diamond) and other entities that share some of their properties (i.e. concrete). The resulting riddles are composed as a sequence of comparisons using this template: “What is … as attribute as concept?”, where “attribute” is a property of the searched concept that is the answer to the riddle, and “concept” is a different entity that shares the value of the attribute with the target concept. For example, “What is … as hard as concrete and as transparent as hair?” is a riddle generated for the concept diamond by the riddle generator. Later, the relation between words associations already available in different web resources and the type of rhetorical figures that people use regularly is analysed. The aim is to identify regularities or heuristics that can be used in a posterior stage to develop a text generator. The capacity of this system will be studied for the automatic generation of analogies, similes and metaphors.
Description
Máster en Ingeniería Informática, Facultad de Informática, Departamento de Ingeniería del Software e Inteligencia Artificial curso 2016-2017