Search

Paraphrase detection is an important unresolved task in natural language processing; especially in the Spanish language. In order to address this issue, and contribute to the creation of high-performance paraphrase detection automated systems, we propose a shared task called PAR-MEX. For this task, we created a corpus, in Spanish, with topics in the domain of Mexican gastronomy. Afterwards, the participants in this task submitted their classification results on our corpus. In this paper we explain the steps followed for the creation of the corpus, we summarize the results obtained by the various participants, and propose some conclusions regarding the paraphrase-detection task in Spanish. La detección de paráfrasis es una tarea importante no resuelta en procesamiento del lenguaje natural; especialmente en la lengua española. Para atacar este problema, y para contribuir a la creación de sistemas de detección automática que obtengan resultados competitivos, proponemos la tarea compartida llamada PAR-MEX. Para esto, creamos un corpus en español con temas dentro del campo semántico de gastronomía mexicana. Después los participantes en esta tarea enviaron los resultados de sus sistemas de clasificación sobre nuestro corpus. En este paper explicamos los pasos seguidos para la creación del corpus, resumimos los resultados obtenidos por los participantes, y proponemos algunas conclusiones al respecto de la detección de paráfrasis en español. We acknowledge the support of the projects CONACyT CB A1-S-27780, and DGAPA-UNAM PAPIIT references TA400121 and TA101722.

Social networks have modified the way we communicate. It is now possible to talk to a large number of people we have never met. Knowing the traits of a person from what he/she writes has become a new area of computational linguistics called Author Profiling. In this paper, we introduce a method for applying transfer learning to address the gender identification problem, which is a subtask of Author Profiling. Systems that use transfer learning are trained in a large number of tasks and then tested in their ability to learn new tasks. An example is to classify a new image into different possible classes, giving an example of each class. This differs from the traditional approach of standard machine learning techniques, which are trained in a single task and are evaluated in new examples of that task. The aim is to train a gender identification model on Twitter users using only their text samples in Spanish. The difference with other related works consists in the evaluation of different preprocessing techniques so that the transfer learning-based fine-tuning is more efficient.

This paper presents the overview of FakeDeS 2021, the second edition of this lab under the IberLEF conference. The FakeDeS shared task aims to explore different methodologies and strategies related to fake news detection in Spanish. This year edition brings two main challenges: thematic and language variation. For this purpose, we introduce a new testing corpus containing news related to COVID-19 and news from other Ibero-American countries. Este artículo hace una presentación general de la tarea compartida FakeDeS 2021, cuya segunda edición ha tenido lugar en 2021 bajo el congreso IberLEF, aunque se trata de la primera vez con esta denominación. La tarea FakeDeS tiene por objetivo explorar diferentes métodos y estategias relacinados con la detección de noticias falsas en español, principalmente en su variante de México. La edición de este año propone dos desafíos principales: variación temática y variación lingüística. Para ello, se introduce un nuevo corpus de prueba que contiene noticias relacionadas con COVID 19 y noticias de otros países de Iber-América. This research was funded by CONACyT project CB A1-S-27780, DGAPA-UNAM PAPIIT grants number TA400121 and TA100520. The authors also thank CONACYT for the computer resources provided through the INAOE Supercomputing Laboratory's Deep Learning Platform for Language Technologies.

The book collects contributions from well-established researchers at the interface between language and cognition. It provides an overview of the latest insights into this interdisciplinary field from the perspectives of natural language processing, computer science, psycholinguistics and cognitive science. One of the pioneers in cognitive natural language processing is Michael Zock, to whom this volume is dedicated. The structure of the book reflects his main research interests: lexicon and lexical analysis, semantics, language and speech generation, reading and writing technologies, language resources and language engineering. The book is a valuable reference work and authoritative information source, giving an overview on the field and describing the state of the art as well as future developments. It is intended for researchers and advanced students interested in the subject.

[Abstract] With the discovery of the DNA structure (Watson and Crick, 1953), the idea of DNA as a linguistic code arose (Monod, 1970). Many researchers have considered DNA as a language, pointing out the semiotic parallelism between genetic code and natural language. This idea had been discussed, almost dismissed and somehow accepted. This paper does not claim that the genetic code is a linguistic structure, but it highlights several important semiotic analogies between DNA and verbal language. Genetic code and natural language share a number of units, structures and operations. The syntactic and semantic parallelisms between those codes should lead to a methodological exchange between biology, linguistics and semiotics. During the 20th century, biology has become a pilot science, so that many disciplines have formulated their theories under models taken from biology. Computer science has become almost a bioinspired field thanks to the great development of natural computing and DNA computing. Biology and semiotics are two different sciences challenged by the same common goal of deciphering the codes of the nature. Linguistics could become another «bio-inspired» science by taking advantage of the structural and «semantic» similarities between the genetic code and natural language. Biological methods coming from computer science can be very useful in the field of linguistics, since they provide flexible and intuitive tools for describing natural languages. In this way, we obtain a theoretical framework where biology, linguistics and computer science exchange methods and interact, thanks to the semiotic parallelism between the genetic code a natural language. The influence of the semiotics of the genetic code in linguistics is parallel to the need of achieving an implementable formal description of natural language. In this paper we present an overview of different bio-inspired methods — from theoretical computer science — that during the last years have been successfully applied to several linguistics issues, from syntax to pragmatics.

SISTEMAS DE INSERCIÓN Y BORRADO: COMPLEJIDAD Y CAPACIDAD DE MODELADO El objetivo central de la tesis es el estudio de los sistemas de inserción y borrado y su capacidad computacional. Más concretamente, estudiamos algunos modelos de generación de lenguaje que usan operaciones de reescritura de dos cadenas. También consideramos una variante distribuida de los sistemas de inserción y borrado en el sentido de que las reglas se separan entre un número finito de nodos de un grafo. Estos sistemas se denominan sistemas controlados mediante grafo, y aparecen en muchas áreas de la Informática, jugando un papel muy importante en los lenguajes formales, la lingüística y la bio-informática. Estudiamos la decidibilidad/ universalidad de nuestros modelos mediante la variación de los parámetros de tamaño del vector. Concretamente, damos respuesta a la cuestión más importante concerniente a la expresividad de la capacidad computacional: si nuestro modelo es equivalente a una máquina de Turing o no. Abordamos sistemáticamente las cuestiones sobre los tamaños mínimos de los sistemas con y sin control de grafo., COMPLEXITY AND MODELING POWER OF INSERTION-DELETION SYSTEMS The central object of the thesis are insertion-deletion systems and their computational power. More specifically, we study language generating models that use two string rewriting operations: contextual insertion and contextual deletion, and their extensions. We also consider a distributed variant of insertion-deletion systems in the sense that rules are separated among a finite number of nodes of a graph. Such systems are refereed as graph-controlled systems. These systems appear in many areas of Computer Science and they play an important role in formal languages, linguistics, and bio-informatics. We vary the parameters of the vector of size of insertion-deletion systems and we study decidability/universality of obtained models. More precisely, we answer the most important questions regarding the expressiveness of the computational model: whether our model is Turing equivalent or not. We systematically approach the questions about the minimal sizes of the insertiondeletion systems with and without the graph-control.

Different types of regulated grammars have been introduced in order to supplement shortcomings of context-free grammars in applications preserving their elegant mathematical properties. However, the rapid developments in present day industry, biology, and other areas challenge to deal with various tasks which need suitable tools for their modelling and investigation. We propose Petri net controlled grammars as models for representing and analyzing of metabolic pathways in living cells where Petri nets are responsible for the structure and communication of the pathways, and grammars represent biochemical processes. On the other hand, the control by Petri nets has also theoretical interest: it extends possibilities to introduce and investigate concurrent control mechanisms in formal language theory. The thesis introduces various variants of Petri net controlled grammars using different types of Petri nets and investigates their mathematical properties such as computational power and closure properties., Los diferentes tipos de gramáticas con reescritura regulada han sido introducidas para complementar las deficiencias de las gramáticas libres del contexto en las aplicaciones, preservando sus propiedades matemáticas. Por otro lado, la rápida evolución la biología, y otras áreas actuales supone un reto para tratar de las tareas varias que necesitan las herramientas adecuadas para la elaboración de modelos e investigación. Proponemos gramáticas controladas por redes de Petri como modelos para representar y analizar los procesos bioquímicos en las células vivas donde redes de Petri son responsables de la estructura, y gramáticas representan los procesos generativos. Además, el control de redes de Petri también tiene interés teórico: amplía las posibilidades de investigar los mecanismos de control concurrente en la teoría de lenguajes formales. La tesis presenta distintas variantes de gramáticas controladas por redes de Petri e investiga sus propiedades matemáticas.

In the field of grammatical inference, the goal of any learning algorithm is to identify a target concept from a given class by having access to a specific type of information. The main learning settings are Gold's model of learning in the limit and Angluin's query learning. This dissertation is primarily concerned with the second approach.We thoroughly investigate a recently introduced, linguistic motivated, type of query called Correction Query (CQ). We consider three possible definitions, and for each of them we give necessary and sufficient conditions for a language class to be learnable with these types of queries. Furthermore, we compare the model of learning with CQs with other well-known Gold-style and query learning models when no efficiency constraints are imposed. Results are also obtained for the restricted version of the model of learning with CQs in polynomial time. Additionally, we discuss the learnability of deterministic finite automata (DFAs) with correction and equivalence queries. We design several learning algorithms and we present a comparison between our algorithms and the standard algorithm for learning DFAs with membership and equivalence queries. These results are furthermore extended from string languages to tree languages., Dentro del ámbito de la inferencia gramatical, el objetivo de cualquier algoritmo de aprendizaje consiste en identificar un concepto de una clase dada, teniendo acceso a un tipo específico de información. Los dos modelos de aprendizaje principales son: el modelo de aprendizaje en el límite propuesto por Gold y el modelo de aprendizaje a partir de preguntas de Angluin. Esta tesis se centra fundamentalmente en el segundo tipo de aprendizaje. En esta tesis doctoral investigamos a fondo un tipo de preguntas introducidas recientemente, por motivos lingüísticos, denominada PREGUNTA DE CORRECCIÓN. Nosotros consideramos tres posibles definiciones, y para cada uno de ellos presentamos condiciones necesarias y suficientes para que una clase de lenguajes sea identificable con estos tipos de preguntas. Además, comparamos estos tres modelos de aprendizaje a partir de preguntas de corrección con otros modelos de aprendizaje, en el caso general (sin ninguna restricción de tiempo) y también imponiendo que los algoritmos sean polinómicos.Además, investigamos el aprendizaje de autómata finitos deterministas (AFD) con preguntas de corrección y equivalencia. Diseñamos varios algoritmos de aprendizaje y presentamos una comparación entre nuestros algoritmos y el algoritmo estándar de aprendizaje a partir de preguntas de pertenencia y equivalencia. Estos resultados son, además, extendidos para los lenguajes de árboles.

Esta tesis está dedicada a los sistemas de membranas con objetos-símbolo como marco teórico de los sistemas paralelos y distribuidos de procesamiento de multiconjuntos.Una computación de parada puede aceptar, generar o procesar un número, un vector o una palabra; por tanto el sistema define globalmente (a través de los resultados de todas sus computaciones) un conjunto de números, de vectores, de palabras (es decir, un lenguaje), o bien una función. En esta tesis estudiamos la capacidad de estos sistemas para resolver problemas particulares, así como su potencia computacional. Por ejemplo, las familias de lenguajes definidas por diversas clases de estos sistemas se comparan con las familias clásicas, esto es, lenguajes regulares, independientes del contexto, generados por sistemas 0L tabulados extendidos, generados por gramáticas matriciales sin chequeo de apariciones, recursivamente enumerables, etc. Se prestará especial atención a la comunicación de objetos entre regiones y a las distintas formas de cooperación entre ellos.Se pretende (Sección 3.4) realizar una formalización los sistemas de membranas y construir una herramienta tipo software para la variante que usa cooperación no distribuida, el navegador de configuraciones, es decir, un simulador, en el cual el usuario selecciona la siguiente configuración entre todas las posibles, estando permitido volver hacia atrás. Se considerarán diversos modelos distribuidos. En el modelo de evolución y comunicación (Capítulo 4) separamos las reglas tipo-reescritura y las reglas de transporte (llamadas symport y antiport). Los sistemas de bombeo de protones (proton pumping, Secciones 4.8, 4.9) constituyen una variante de los sistemas de evolución y comunicación con un modo restrictivo de cooperación. Un modelo especial de computación con membranas es el modelo puramente comunicativo, en el cual los objetos traspasan juntos una membrana. Estudiamos la potencia computacional de las sistemas de membranas con symport/antiport de 2 o 3 objetos (Capítulo 5) y la potencia computacional de las sistemas de membranas con alfabeto limitado (Capítulo 6).El determinismo (Secciones 4.7, 5.5, etc.) es una característica especial (restrictiva) de los sistemas computacionales. Se pondrá especial énfasis en analizar si esta restricción reduce o no la potencia computacional de los mismos. Los resultados obtenidos para sistemas de bombeo del protones están transferidos (Sección 7.3) a sistemas con catalizadores bistabiles. Unos ejemplos de aplicación concreta de los sistemas de membranas (Secciones 7.1, 7.2) son la resolución de problemas NP-completos en tiempo polinomial y la resolución de problemas de ordenación., This thesis deals with membrane systems with symbol objects as a theoretical framework of distributed parallel multiset processing systems.A halting computation can accept, generate or process a number, a vector or a word, so the system globally defines (by the results of all its computations) a set of numbers or a set of vectors or a set of words, (i.e., a language), or a function. The ability of these systems to solve particular problems is investigated, as well as their computational power, e.g., the language families defined by different classes of these systems are compared to the classical ones, i.e., regular, context-free, languages generated by extended tabled 0L systems, languages generated by matrix grammars without appearance checking, recursively enumerable languages, etc. Special attention is paid to communication of objects between the regions and to the ways of cooperation between the objects.An attempt to formalize the membrane systems is made (Section 3.4), and a software tool is constructed for the non-distributed cooperative variant, the configuration browser, i.e., a simulator, where the user chooses the next configuration among the possible ones and can go back. Different distributed models are considered. In the evolution-communication model (Chapter 4) rewriting-like rules are separated from transport rules. Proton pumping systems (Sections 4.8, 4.9) are a variant of the evolution-communication systems with a restricted way of cooperation. A special membrane computing model is a purely communicative one: the objects are moved together through a membrane. We study the computational power of membrane systems with symport/antiport of 2 or 3 objects (Chapter 5) and the computational power of membrane systems with a limited alphabet (Chapter 6).Determinism (Sections 4.7, 5.5, etc.) is a special property of computational systems; the question of whether this restriction reduces the computational power is addressed. The results on proton pumping systems can be carried over (Section 7.3) to the systems with bi-stable catalysts. Some particular examples of membrane systems applications are solving NP-complete problems in polynomial time, and solving the sorting problem.