Use this link to cite:
http://hdl.handle.net/2183/19278 Segmentación de palabras en español mediante modelos del lenguaje basados en redes neuronales
Loading...
Identifiers
Publication date
Authors
Advisors
Other responsabilities
Journal Title
Bibliographic citation
Yerai Doval, Carlos Gómez-Rodríguez and Jesús Vilares, Segmentación de palabras en español mediante modelos del lenguaje basados en redes neuronales, Procesamiento del Lenguaje Natural, 57:75-82, 2016
Type of academic work
Academic degree
Abstract
[Resumen] En las plataformas de microblogging abundan ciertos tokens especiales como los hashtags o las menciones en los que un grupo de palabras se escriben juntas sin espaciado entre ellas; p.ej.: #añobisiesto o @ryanreynoldsnet. Debido a la forma en que se escriben este tipo de textos, este fenómeno de ensamblado de palabras puede aparecer junto a su opuesto, la segmentación de palabras, afectando a cualquier elemento del texto y dificultando su análisis. En este trabajo se muestra un enfoque algorítmico que utiliza como base un modelo del lenguaje —en nuestro caso concreto uno basado en redes euronales— para resolver el problema de la segmentación y ensamblado de palabras, en el que se trata de recuperar el espaciado estándar de las palabras que han sufrido alguna de estas transformaciones añadiendo o quitando espacios donde corresponda. Los resultados obtenidos son prometedores e indican que tras un mayor refinamiento del modelo del lenguaje se podrá sobrepasar al estado del arte.
[Abstract] In social media platforms special tokens abound such as hashtags and mentions in which multiple words are written together without spacing between them; e.g. #leapyear or @ryanreynoldsnet. Due to the way this kind of texts are written, this word assembly phenomenon can appear with its opposite, word segmentation, a ecting any token of the text and making it more di cult to perform analysis on them. In this work we show an algorithmic approach based on a language model |in this case a neural model| to solve the problem of the segmentation and assembly of words, in which we try to recover the standard spacing of the words that have su ered one of these transformations by adding or deleting spaces when necessary. The promising results indicate that after some further re nement of the language model it will be possible to surpass the state of the art.
[Abstract] In social media platforms special tokens abound such as hashtags and mentions in which multiple words are written together without spacing between them; e.g. #leapyear or @ryanreynoldsnet. Due to the way this kind of texts are written, this word assembly phenomenon can appear with its opposite, word segmentation, a ecting any token of the text and making it more di cult to perform analysis on them. In this work we show an algorithmic approach based on a language model |in this case a neural model| to solve the problem of the segmentation and assembly of words, in which we try to recover the standard spacing of the words that have su ered one of these transformations by adding or deleting spaces when necessary. The promising results indicate that after some further re nement of the language model it will be possible to surpass the state of the art.







