• Digital Humanities and Cultural Heritage
• UPCommons. Portal del coneixement o... close

El procesamiento de lenguaje natural, o NLP (Natural Language Processing), a lo largo de los últimos años ha ido adquiriendo cada vez más importancia en el mundo de la tecnología. Incluso las grandes empresas como Google, Apple o Microsoft ya han hecho sus propios productos sobre este tema. No obstante, el estudio del NLP es un mundo bastante amplio, empezando por sus finalidades, las cuales pueden ser muy diferentes: traducción de idiomas a tiempo real, incorporarlo en proyectos de domótica, etc. Hasta en las herramientas que se utilizan para trabajar sobre este, en nuestro proyecto nos hemos querido centrar en las redes de transición aumentadas, o ATN (Augmented Transition Network). Hoy día la implementación de un ATN es un poco tediosa, por el hecho de que los lenguajes de programación más utilizados no facilitan las herramientas adecuadas. Por eso hemos querido confeccionar nuestro propio lenguaje, enfocado concretamente en la realización de ATN's. De manera que la implementación de un ATN sea más simple, i de esta manera sea más viable a la hora de incorporarlos en proyectos más grandes sobre NLP. El processament de llenguatge natural, o NLP (Natural Language Processing), al llarg dels últims anys ha anat adquirint cada cop més importància en el món de la tecnologia. Fins i tot grans empreses com Google, Apple o Microsoft ja han fet els seus propis productes sobre aquest tema. No obstant això, l'estudi del NLP és un món força ampli, començant per les seves finalitats, les quals poden ser molt diferents: traducció d'idiomes a temps real, incorporar-ho en projectes de domòtica, etc. Fins en les eines que s'utilitzen per a treballar-hi, en el nostre projecte ens hem volgut centrar en les xarxes de transició augmentades, o ATN (Augmented Transition Network). Avui dia la implementació d'un ATN és una mica tediosa, pel fet que els llenguatges de programació més utilitzats no faciliten les eines adients. Per això hem confeccionat el nostre propi llenguatge, enfocat concretament en la realització de ATN's. De manera que la implementació d'un ATN sigui més senzilla, i d'aquesta manera sigui més viable a l'hora d'incorporar-los en projectes més grans sobre NLP. The natural language processing, or NLP, has become increasingly important in the world of technology over the last few years. Even large companies like Google, Apple or Microsoft have already made their own products on this subject. However, the study of the NLP is a very broad world, starting with its purposes, which can be very different: translation of languages in real time, incorporate it into domotic projects, etc. Even in the tools used to work on this, in our project we wanted to focus on the augmented transition network, or ATN. Today the implementation of an ATN is a bit tedious, due to the fact that the most used programming languages do not provide the right tools. That's why we wanted to create our own language, focused specifically on the realization of ATN's. So that the implementation of an ATN is simpler, and in this way is more viable when incorporating them into larger NLP projects.

With the expansion of the communication platforms, individuals are very used to exchange information and communicate with other people through these platforms, both for social and business purposes. It’s a known problem, that many people use the anonymity provided by these platforms to use inappropriate and offensive language. The company NextreT S.L. has built a communication platform directed to business use. As the company wants to avoid offensive language in this platform, natural language processing tools in big data environment are going to be used to analyze each written text to detect and remove if required this inappropriate and offensive language. During this project, a variety of techniques used in the State of the Art are analyzed, compared and then tested using a completely new data set in Spanish language created using Tellfy App and Twitter corpus. Initially, different word encoding methods are tested, including word embedding like Word2Vec and FastText. In addition, different hyper parameter configurations are checked as well as model performances with different data sizes. Finally, after a forward feature selection phase, model ensemble techniques are tested. During these tests, it has been shown that the combination of the features that are used is very important to increase the performance of the models. Also, the different word representation techniques are very related to the performance of the models. Furthermore, the sizes of the training sets that are used need to be as representative and as large as possible. Finally, after using different complex Deep Neural Network models, more traditional Logistic Regression models can offer a better performance.

Question generation attempts to generate a natural language question given a passage and an answer. Most state-of-the-art methods have focused on generating simple questions involving single-hop relations and based on a single or a few sentences. In this project, we focus on generating multi-hop questions which requires discovering and modeling the multiple entities and their semantic relations in the passage. To that end, we use the HotpotQA dataset, a multi-document and multi-hop dataset for questions answering that provides not only the context, question, and answer but also the supporting facts that lead to the answer. To solve the problem, we propose the use of transformer-based models, which have shown to perform well in single-hop question generation, and we study different variants to condition the model using the context and the supporting facts.

Process models are a fundamental element in the BPM lifecycle. Hence, it is of paramount importance for organizations to rely on high-quality, accurate and up-to-date process models, to avoid taking decisions on the basis of a wrong picture of the reality. In this demo we present modeljudge.cs.upc.edu, a platform to boost the training of novice modelers when confronted with the task of translating a textual description into a process model in BPMN notation. The platform is integrated with Natural Language Processing (NLP) analysis and textual annotation, together with a novel model-to-text alignment technique. By using this platform, a novice modeler will receive diagnostics in real-time, which may contribute to a more satisfactory modeling experience. Peer Reviewed

A semantic tagger aiming to detect relevant entities in Arabic medical documents and tagging them with their appropriate semantic class is presented. The system takes profit of a Multilingual Framework covering four languages (Arabic, English, French, and Spanish), in a way that resources available for each language can be used to improve the results of the others, this is specially important for less resourced languages as Arabic. The approach has been evaluated against Wikipedia pages of the four languages belonging to the medical domain. The core of the system is the definition of a base tagset consisting of the three most represented classes in SNOMED-CT taxonomy and the learning of a binary classifier for each semantic category in the tagset and each language, using a distant learning approach over three widely used knowledge resources, namely Wikipedia, Dbpedia, and SNOMED-CT. Peer Reviewed

SKATER Internal Report: software of infrastructure for extractive Summarization (work carried out until December 2013)

Co-reference resolution is an important part of natural language understanding and it's been affected by the current corpora lacking in diversity. This project presents the implementation of two models for masked language modeling using pre-trained BERT adjusted to work for a classification problem.