Antonio Ferrández Rodríguez

Processing of Spanish Definite Descriptions

Natural language processing systems require a suitable linguistic co-reference resolution. Many works have been focused on pronominal or adjective anaphora resolution in English and Spanish texts. However, few of them have been focused on the resolution of definite descriptions. Traditionally, anaphora resolution was not usually treated properly as a linguistic phenomenon, due fundamentally to the lack of linguistic knowledge or suitable sources of information for its resolution. In this work, a method for the resolution of the identity co-reference produced by definite descriptions in Spanish texts is presented. This method is based on the linguistic knowledge acquired by POS-tagger, synonymous dictionary and relationships between names and verbs resources. This method uses a system of restrictions and preferences in order to obtain the correct antecedent. The method achieves a precision of 95% for the identification of non-anaphoric and anaphoric uses of definite descriptions. Moreover, the method achieves a precision of 80% for the reference resolution task. In addition, a profound study of linguistic characteristics of definite descriptions in a Spanish corpus has been made. From this study, a classification of the different definite description types has been obtained.

Text Segmentation for Efficient Information Retrieval

Previous works in Information Retrieval show that using pieces of text obtain better results than using the whole document as the basic unit to compare with the users query. This kind of IR systems is usually called Passage Retrieval (PR). However, there is not a general agreement about how one should define those pieces of text (also known as passages), in order to obtain an optimum performance. This paper proposes a PR system based on a novel selection of variable size passages. It presents an evaluation that shows better results than a standard IR system and several well-known PR systems.

Generation of Spanish Zero-Pronouns into English

It is widely agreed that anaphora and ellipsis resolution is an important problem that is still to be solved in Natural Language Processing systems, such as Machine Translation and Information Extraction applications. Zero-pronouns are a special kind of anaphora, whose resolution also lies in ellipsis phenomenon since they do not appear explicitly in the text. They must first be detected (ellipsis), and then resolved just like any other pronoun (anaphora). This kind of pronoun occurs in Spanish texts when they occupy the grammatical position of the subject. In this paper, we propose an approach that resolves zero-pronouns in Spanish texts and subsequently generates them into English. A success rate of 75% has been obtained in the generation of Spanish zero-pronouns into English.

Using a Passage Retrieval System to Support Question Answering Process

Previous works in Information Retrieval show that using pieces of text obtain better results than using the whole document as the basic unit to compare with the users query. This kind of IR systems is usually called Passage Retrieval (PR). This paper discusses the use of our PR system in the question answering process (QA). Our main objective is to examine if a PR system provide a better support for QA than Information Retrieval systems based in the whole document.

Pronominal Anaphora Generation in an English-Spanish MT Approach

This paper presents the pronominal anaphora generation module in a Machine Translation (MT) system. The MT interlingua approach -AGIR (Anaphora Generation with an Interlingua Representation)- allows the generation of anaphoric expressions into the target language from the interlingua representation of the source text.AGIR uses different kinds of knowledge (lexical, syntactic, morphological and semantic information) to solve the Natural Language Processing (NLP) problems of the source text. Subsequently, an interlingua representation of the whole text is obtained that allows the correct generation of anaphoric expressions. In this paper we have evaluated the generation of English and Spanish (including zero pronouns) third person personal pronouns into the target language. We have obtained the following results: a precision of 80.39% and 84.77% in the generation of Spanish and English pronominal anaphora respectively.

Using Long Queries in a Passage Retrieval System

Previous works in Information Retrieval show that using pieces of text obtain better results than using the whole document as the basic unit to compare with the users query. This kind of IR systems is usually called Passage Retrieval (PR). However, there is not a general agreement about how one should define those pieces of text (also known as passages), in order to obtain an optimum performance. This paper proposes a PR system based on a novel selection of variable size passages. It presents an evaluation that shows better results than a standard IR system and several well-known PR systems. Also, the influence of short queries with reference to long queries is evaluated.

IL MT System. Evaluation for Spanish-English Pronominal Anaphora Generation

In this paper the pronominal anaphora generation module of a complete interlingua Machine Translation (MT) approach is presented. The approach named AGIR (Anaphora Generation with an Interlingua Representation) allows the generation of anaphoric expressions into the target language from the interlingua representation of the source text. AGIR uses different kinds of knowledge (lexical, syntactic, morphological and semantic information) to solve the Natural Language Processing (NLP) problems of the source text. The paper presents the evaluation of the generation of English and Spanish (including zero pronouns) third person personal pronouns into the target language. The following results have been obtained: a precision of 80.39% and 84.77% in the generation of Spanish and English pronominal anaphora respectively.