Despite significant investments in the normalization and standardization of Electronic Health Records (EHRs), free text is still rule rather than exception clinical notes. The use has implications data reuse methods used for supporting research since query mechanisms cohort definition patient matching are mainly based on structured terminologies. This study aims to develop a method secondary by: (a) using Natural Language Processing (NLP) tagging notes with biomedical terminology; (b)...

This paper describes the participation of ELiRF-UPV team at task 4 SemEval2017. Our approach is based on use convolutional and recurrent neural networks combination general specific word embeddings with polarity lexicons. We participated in all proposed subtasks both for English Arabic languages using same system small variations.

Emotions are central to understanding contemporary journalism; however, they overlooked in automatic news summarization. Actually, summaries an entry point the source article that could favor some emotions captivate reader. Nevertheless, emotional content of summarization corpora and behavior models still unexplored. In this work, we explore usage established methodologies study models. Using these methodologies, two widely used corpora: Cnn/Dailymail Xsum, capabilities three...

This paper describes the participation of ELiRF-UPV team at tasks 1 and 3 Semeval-2018. We present a deep learning based system that assembles Convolutional Neural Networks Long Short-Term Memory neural networks. has been used with slight modifications for two addressed both English Spanish. Finally, results obtained in competition are reported discussed.

We present an approach for the development of Language Understanding systems from a Transduction point view. describe use two types automatically inferred transducers as appropriate models understanding phase in dialog systems.

In this article, we present an approach to the development of a stochastic dialog manager. The model used by manager generate its turns takes into account both last user and system, information supplied throughout dialog. As space situations that can be presented in dialogs is too large, some techniques for reducing have been proposed. This system has developed DIHANA project, whose goal design access railway using spontaneous speech Spanish. A training corpus 900 dialogs, was acquired...

We are interested in the problem of learning Spoken Language Understanding (SLU) models for multiple target languages.Learning such requires annotated corpora, and porting to different languages would require corpora with parallel text translation semantic annotations.In this paper we investigate how learn a SLU model language starting from no annotation.Our proposed algorithm is based on idea exploiting diversity (with regard performance coverage) systems transfer statistically stable...

Question answering (QA) is probably one of the most challenging tasks in field natural language processing. It requires search engines that are capable extracting concise, precise fragments text contain an answer to a question posed by user. The incorporation voice interfaces QA systems adds more and very appealing perspective for these systems. This paper provides comprehensive description current state-of-the-art voice-activated Finally, scenarios will emerge from introduction speech...

In this paper, we present an approach to spoken dialog management based on the use of a Stochastic Finite-State Transducer estimated from corpus. The states represent states, input alphabet includes all possible user utterances, without considering specific values, and set system answers constitutes output alphabet. Then, describes path in transducer model initial state final one. An automatic generation technique was used order generate corpus which parameters are estimated. Our proposal...

This work is partially supported by the Spanish MICINN under contract TIN2011-28169-C05-01.

Social media has led to a redefinition of the journalist’s role. Specifically on Twitter, these professionals assume an influential position and their discourse is dominated by personal opinions. Taking into consideration that this platform proven be breeding ground for polarization, digital harassment hate speech, notably against women politicians, research aims analyze journalists’ involvement in complex scenario. The investigation determine whether, immersed online gender defamation...

This paper describes our proposal for Sentiment Analysis in Twitter the Spanish language. The main characteristics of system are use word embedding specifically trained from tweets and self-attention mechanisms that allow to consider sequences without using convolutional nor recurrent layers. These based on encoders Transformer model. results obtained Task 1 TASS 2019 workshop, all variants proposed, support correctness adequacy proposal.

Most of the models proposed in literature for abstractive summarization are generally suitable English language but not other languages. Multilingual were introduced to address that constraint, despite their applicability being broader than monolingual models, performance is typically lower, especially minority languages like Catalan. In this paper, we present a model textual content Catalan language. The Transformer encoder-decoder which pretrained and fine-tuned specifically using corpus...

In this work, a general theoretical framework for extractive summarization is proposed—the Attentional Extractive Summarization framework. Although abstractive approaches are generally used in text today, methods can be especially suitable some applications, and they help with other tasks such as Text Classification, Question Answering, Information Extraction. The proposed approach based on the interpretation of attention mechanisms hierarchical neural networks, which compute document-level...

In this paper we propose an algorithm to learn statistical language understanding models from a corpus of unaligned pairs sentences and their corresponding semantic representation. Specifically, it allows automatically map variablelength word segments with units thus, the decoding user utterances meanings. way avoid time consuming work manually associate labels words, process which is needed by almost all corpus-based approaches. We use component Spoken Dialog System for railway information...

This paper describes our participation at tasks 10 (sub-task B, Message Polarity Classification) and 11 task (Sentiment Analysis of Figurative Language in Twitter) Semeval2015.We describe the Support Vector Machine system we used this competition.We also present relevant feature set that take into account models.Finally, show results obtained competition some conclusions.