Abstract. This paper describes a multiagent system for natural lan-guage processing that deals wi... more Abstract. This paper describes a multiagent system for natural lan-guage processing that deals with task oriented dialogue processing in Portuguese. The system is based on Discourse Structure Theory, on Centering Theory and on Lochbaum's work on modeling the Intentional ...
Natural language inference is the task of automatically identifying whether a given text (premise... more Natural language inference is the task of automatically identifying whether a given text (premise) implies another (hypothesis). Among multiple possible applications, it is especially relevant in the legal field to understand textual entailment between legal sentences, being the focus of recent research efforts. In this work, we evaluated the usage of BERT for natural language inference by conducting experiments and comparing results obtained by testing on a larger corpus with texts from multiple domains and a smaller corpus of legal sentences. Furthermore, we conducted a cross-experiment by training on the larger corpus and testing on the legal corpus. As a result, we obtained a mean accuracy of 88.91% in the corpus with multiple domains, a value comparable to related work. However, the same technique presented lower scores in the legal corpus and the cross-experiment.
English. In this article, we present the results of applying a Stacking Ensemble method to the pr... more English. In this article, we present the results of applying a Stacking Ensemble method to the problem of hate speech classification proposed in the main task of HaSpeeDe 2 at EVALITA 2020. The model was then compared to a Logistic Regression classifier, along with two other benchmarks defined by the competition's organising committee (an SVM with a linear kernel and a majority class classifier). Results showed our Ensemble to outperform the benchmarks to various degrees, both when testing in the same domain as training and in a different domain. 1 Introduction Social networks are already part of people's lives, generating thousands of publications on a daily basis. Even though most of this material presents no
Pronominal anaphora resolution consists in finding a referent for a given pronoun. Although being... more Pronominal anaphora resolution consists in finding a referent for a given pronoun. Although being essential for many natural language processing systems, such as automatic translators, text generators and summarizers, there is a myriad of issues regarding this task, particularly when there is more than one possible referent for a given pronoun. Over the years, several approaches have been proposed to deal with this challenge, usually taking only syntactic information into account. On the other hand, methods based on Centering Theory rely on concepts such as coherence, for instance, to do the job. In this work, we describe our implementation and evaluation of existing centering-based algorithms for pronominal resolution in Portuguese. As a result, we indicate both the strong and weak points of each of the tested algorithms, thereby helping other researchers to make a more informed decision about which method to use.
Distributional models have become popular due to the abstractions that allowed their immediate us... more Distributional models have become popular due to the abstractions that allowed their immediate use, with good results and little implementation effort when compared to precursor models. Given their presumed high level of generalization it would be expected that good and similar results would be found in data sets sharing the same nature and purpose. However, this is not always the case. In this work, we present the results of the application of BERTimbau in two related data sets, built for the task of Semantic Similarity identification, with the goal of detecting redundancy in text. Results showed that there are considerable differences in accuracy between the data sets. We explore aspects of the data sets that could explain why accuracy results are different across them.
English. In this article, we present the results of applying a Stacking Ensemble method to the pr... more English. In this article, we present the results of applying a Stacking Ensemble method to the problem of hate speech classification proposed in the main task of HaSpeeDe 2 at EVALITA 2020. The model was then compared to a Logistic Regression classifier, along with two other benchmarks defined by the competition’s organising committee (an SVM with a linear kernel and a majority class classifier). Results showed our Ensemble to outperform the benchmarks to various degrees, both when testing in the same domain as training and in a different domain. Italiano. In questo articolo, ci presentiamo i risultati dell’applicazione di un modello di Stacking Ensemble al problema della classificazione dei discorsi di incitamento all’odio nel compito A di EVALITA (HaSpeeDe 2). Il modello è stato quindi confrontato con un modello di regressione logistica, insieme ad altri due benchmark definiti dal comitato organizzatore della competizione (un SVM con un kernel lineare e un classificatore di classe...
This paper describes the use of neca's rrl-Rich Representation Language-in the aid of automatic d... more This paper describes the use of neca's rrl-Rich Representation Language-in the aid of automatic dialogue summarisation. We start by defining a set of summary acts, and then present a mapping between these acts, taken from human generated summaries, and their corresponding speech acts from the source dialogues. This mapping can then be used by a summary planner to identify portions in the dialogue with higher probability to support some argument it is about to introduce in the summary. Depending on the application, it could also be used to gather, in the source dialogue, specific information to fulfill summary templates. In following this procedure, we hope researchers will be able to build more human-like summaries, by linking corpora summaries to their source dialogues.
Anais do XIII Simpósio Brasileiro de Tecnologia da Informação e da Linguagem Humana (STIL 2021), 2021
This paper presents the project of a large multi-genre treebank for Brazilian Portuguese, called ... more This paper presents the project of a large multi-genre treebank for Brazilian Portuguese, called Porttinari. We address relevant research questions in its construction and annotation, reporting the work already done. The treebank is affiliated with the “Universal Dependencies” international model, widely adopted in the area, and must be the basis for the development of state of the art tagging and parsing systems for Portuguese, as well as for conducting linguistic studies on morphosyntax and syntax for this language.
Anais do Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), Nov 29, 2021
Automatically dealing with Natural Language User-Generated Content (UGC) is a challenging task of... more Automatically dealing with Natural Language User-Generated Content (UGC) is a challenging task of utmost importance, given the amount of information available over the web. We present in this paper an effort on building tokenization and Part of Speech (PoS) tagging systems for tweets in Brazilian Portuguese, following the guidelines of the Universal Dependencies (UD) project. We propose a rule-based tokenizer and the customization of current state-of-the-art UD-based tagging strategies for Portuguese, achieving a 98% f-score for tokenization, and a 95% f-score for PoS tagging. We also introduce DANTEStocks, the corpus of stock market tweets on which we base our work, presenting preliminary evidence of the multi-genre capacity of our PoS tagger.
This presentation introduces the project of a large multi-genre treebank for Brazilian Portuguese... more This presentation introduces the project of a large multi-genre treebank for Brazilian Portuguese, called Porttinari. We address relevant research questions in its construction and annotation, reporting the work already done. The treebank is affiliated with the "Universal Dependencies" international model, widely adopted in the area, and must be the basis for the development of state of the art tagging and parsing systems for Portuguese, as well as for conducting linguistic studies on morphosyntax and syntax for this language.
EVALITA Evaluation of NLP and Speech Tools for Italian - December 17th, 2020, 2020
English. In this article, we describe two classification models (a Convolutional Neural Network a... more English. In this article, we describe two classification models (a Convolutional Neural Network and a Logistic Regression classifier), arranged according to three different strategies, submitted to subtask A of Automatic Misogyny Identification at EVALITA 2020. Results were very encouraging for detecting misogyny, even though aggressiveness was less accurate. Our second strategy, consisting of a Convolutional Neural Network and logistic regression to identify misogyny and aggressiveness, respectively, won the sixth place in the competition. Italiano. In questo articolo, descriviamo due modelli di classificazione (i.e., Convolutional Neural Network e Regressione Logistica), organizzati secondo tre diverse strategie, per il subtask A dello shared task Automatic Misogyny Identification a EVALITA 2020. I risultati sono stati molto incoraggianti nel rilevamento della misoginia, anche se l'aggressività viene riconosciuta con una precisione più basse. La nostra seconda strategia (Convolutional Neural Network per misoginia e Regressione Logistica per aggressività) ci ha permesso di ottenere il sesto posto nella competizione.
Anais do XXXII Simpósio Brasileiro de Informática na Educação (SBIE 2021), 2021
In this article, we present a systematic literature review, carried out from February to March 20... more In this article, we present a systematic literature review, carried out from February to March 2020, on the application of a machine learning technique to predict student dropout in higher education institutions. Besides describing the protocol followed during our research, which includes the research questions, searched databases and query strings, along with criteria for inclusion and exclusion of articles, we also present our main results, in terms of the attributes used by current research on this theme, along with adopted approaches, specific algorithms, and evalution metrics. The Decision Tree technique is the most used for the construction of models, and accuracy and recall and precision being the most used metric for evaluating models.
Anais do XIII Simpósio Brasileiro de Tecnologia da Informação e da Linguagem Humana (STIL 2021), 2021
A anotação de textos informais segundo a Universal Dependencies requer dois processos de segmenta... more A anotação de textos informais segundo a Universal Dependencies requer dois processos de segmentação: delimitação da unidade relevante para a análise sintática e identificação das palavras sintáticas. Neste artigo, apresentam-se as idiossincrasias linguísticas do corpus DANTEStocks, composto por tweets do mercado financeiro, escritos em Português, e as estratégias gerais de segmentação automática. Assim, contribui-se para a descrição de aspectos linguísticos dos tweets e para o desenvolvimento de recursos e ferramentas de processamento automático desse subgênero de "user-generated content".
Stock price forecasting is a non-linear, dynamic, complex, non-parametric and chaotic problem, be... more Stock price forecasting is a non-linear, dynamic, complex, non-parametric and chaotic problem, being one of the biggest challenges both in financial, academic, and technological terms, especially in what concerns decision support tools. In fact, only few investors are capable of delivering above-market performances in the long run. Within this context, this project aims at modelling a Multi-layer Perceptron Artificial Neural Network (MLP-ANN) capable of forecasting IBOVESPA’s closure trend, based on closure trends of a set of international indexes. Starting from a set of twelve indexes, from three continents, we modelled ANNs that take as input the closure trend of one index per continent, classifying IBOVESPA’s next closure as higher or lower than the closure in the day before. Along with different input sets, ANNs were tested with different training parameters. We then compare the results with those by (Malagrino, 2012), in which the same data set was used to train a Bayesian Netw...
DG.O (Inter)National Conference on Digital Government Research, 2009
With the growing importance of foreign commerce comes also greater opportunities for fraudulent b... more With the growing importance of foreign commerce comes also greater opportunities for fraudulent behaviour. As such, governments must try to detect frauds as soon as they take place, if they are to avoid the profound damage to the so- ciety frauds may cause. Although current fraud detection systems can be used on this endeavour with reasonable ac- curacy, they still
DG.O (Inter)National Conference on Digital Government Research, 2008
There is an increasing concern about the control of customs operations. While globalization incen... more There is an increasing concern about the control of customs operations. While globalization incentives the opening of the market, increasing amounts of imports and exports have been used to conceal several illicit activities, such as, tax evasion, smuggling, money laundry, and drug tra!c. This fact makes it paramount for governments to find automatic or semi-automatic solutions to guide the customs'
In this paper, we introduce a corpus of human-authored dialogue summaries collected through a web... more In this paper, we introduce a corpus of human-authored dialogue summaries collected through a web-experiment. The corpus features (i) one of the few existing corpora of written dialogue summaries; (ii) the only corpus available for dialogue summaries in Portuguese; and (iii) the only available corpus of summaries produced for dialogues whose participants' politeness alignment was systematically varied. Comprising 1,808 human-authored summaries, produced by 452 summarisers, for four different dialogues, this is, to the best of our knowledge, the largest individual corpus available for dialogue summaries, with the highest number of participants involved.
Abstract. This paper describes a multiagent system for natural lan-guage processing that deals wi... more Abstract. This paper describes a multiagent system for natural lan-guage processing that deals with task oriented dialogue processing in Portuguese. The system is based on Discourse Structure Theory, on Centering Theory and on Lochbaum's work on modeling the Intentional ...
Natural language inference is the task of automatically identifying whether a given text (premise... more Natural language inference is the task of automatically identifying whether a given text (premise) implies another (hypothesis). Among multiple possible applications, it is especially relevant in the legal field to understand textual entailment between legal sentences, being the focus of recent research efforts. In this work, we evaluated the usage of BERT for natural language inference by conducting experiments and comparing results obtained by testing on a larger corpus with texts from multiple domains and a smaller corpus of legal sentences. Furthermore, we conducted a cross-experiment by training on the larger corpus and testing on the legal corpus. As a result, we obtained a mean accuracy of 88.91% in the corpus with multiple domains, a value comparable to related work. However, the same technique presented lower scores in the legal corpus and the cross-experiment.
English. In this article, we present the results of applying a Stacking Ensemble method to the pr... more English. In this article, we present the results of applying a Stacking Ensemble method to the problem of hate speech classification proposed in the main task of HaSpeeDe 2 at EVALITA 2020. The model was then compared to a Logistic Regression classifier, along with two other benchmarks defined by the competition's organising committee (an SVM with a linear kernel and a majority class classifier). Results showed our Ensemble to outperform the benchmarks to various degrees, both when testing in the same domain as training and in a different domain. 1 Introduction Social networks are already part of people's lives, generating thousands of publications on a daily basis. Even though most of this material presents no
Pronominal anaphora resolution consists in finding a referent for a given pronoun. Although being... more Pronominal anaphora resolution consists in finding a referent for a given pronoun. Although being essential for many natural language processing systems, such as automatic translators, text generators and summarizers, there is a myriad of issues regarding this task, particularly when there is more than one possible referent for a given pronoun. Over the years, several approaches have been proposed to deal with this challenge, usually taking only syntactic information into account. On the other hand, methods based on Centering Theory rely on concepts such as coherence, for instance, to do the job. In this work, we describe our implementation and evaluation of existing centering-based algorithms for pronominal resolution in Portuguese. As a result, we indicate both the strong and weak points of each of the tested algorithms, thereby helping other researchers to make a more informed decision about which method to use.
Distributional models have become popular due to the abstractions that allowed their immediate us... more Distributional models have become popular due to the abstractions that allowed their immediate use, with good results and little implementation effort when compared to precursor models. Given their presumed high level of generalization it would be expected that good and similar results would be found in data sets sharing the same nature and purpose. However, this is not always the case. In this work, we present the results of the application of BERTimbau in two related data sets, built for the task of Semantic Similarity identification, with the goal of detecting redundancy in text. Results showed that there are considerable differences in accuracy between the data sets. We explore aspects of the data sets that could explain why accuracy results are different across them.
English. In this article, we present the results of applying a Stacking Ensemble method to the pr... more English. In this article, we present the results of applying a Stacking Ensemble method to the problem of hate speech classification proposed in the main task of HaSpeeDe 2 at EVALITA 2020. The model was then compared to a Logistic Regression classifier, along with two other benchmarks defined by the competition’s organising committee (an SVM with a linear kernel and a majority class classifier). Results showed our Ensemble to outperform the benchmarks to various degrees, both when testing in the same domain as training and in a different domain. Italiano. In questo articolo, ci presentiamo i risultati dell’applicazione di un modello di Stacking Ensemble al problema della classificazione dei discorsi di incitamento all’odio nel compito A di EVALITA (HaSpeeDe 2). Il modello è stato quindi confrontato con un modello di regressione logistica, insieme ad altri due benchmark definiti dal comitato organizzatore della competizione (un SVM con un kernel lineare e un classificatore di classe...
This paper describes the use of neca's rrl-Rich Representation Language-in the aid of automatic d... more This paper describes the use of neca's rrl-Rich Representation Language-in the aid of automatic dialogue summarisation. We start by defining a set of summary acts, and then present a mapping between these acts, taken from human generated summaries, and their corresponding speech acts from the source dialogues. This mapping can then be used by a summary planner to identify portions in the dialogue with higher probability to support some argument it is about to introduce in the summary. Depending on the application, it could also be used to gather, in the source dialogue, specific information to fulfill summary templates. In following this procedure, we hope researchers will be able to build more human-like summaries, by linking corpora summaries to their source dialogues.
Anais do XIII Simpósio Brasileiro de Tecnologia da Informação e da Linguagem Humana (STIL 2021), 2021
This paper presents the project of a large multi-genre treebank for Brazilian Portuguese, called ... more This paper presents the project of a large multi-genre treebank for Brazilian Portuguese, called Porttinari. We address relevant research questions in its construction and annotation, reporting the work already done. The treebank is affiliated with the “Universal Dependencies” international model, widely adopted in the area, and must be the basis for the development of state of the art tagging and parsing systems for Portuguese, as well as for conducting linguistic studies on morphosyntax and syntax for this language.
Anais do Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), Nov 29, 2021
Automatically dealing with Natural Language User-Generated Content (UGC) is a challenging task of... more Automatically dealing with Natural Language User-Generated Content (UGC) is a challenging task of utmost importance, given the amount of information available over the web. We present in this paper an effort on building tokenization and Part of Speech (PoS) tagging systems for tweets in Brazilian Portuguese, following the guidelines of the Universal Dependencies (UD) project. We propose a rule-based tokenizer and the customization of current state-of-the-art UD-based tagging strategies for Portuguese, achieving a 98% f-score for tokenization, and a 95% f-score for PoS tagging. We also introduce DANTEStocks, the corpus of stock market tweets on which we base our work, presenting preliminary evidence of the multi-genre capacity of our PoS tagger.
This presentation introduces the project of a large multi-genre treebank for Brazilian Portuguese... more This presentation introduces the project of a large multi-genre treebank for Brazilian Portuguese, called Porttinari. We address relevant research questions in its construction and annotation, reporting the work already done. The treebank is affiliated with the "Universal Dependencies" international model, widely adopted in the area, and must be the basis for the development of state of the art tagging and parsing systems for Portuguese, as well as for conducting linguistic studies on morphosyntax and syntax for this language.
EVALITA Evaluation of NLP and Speech Tools for Italian - December 17th, 2020, 2020
English. In this article, we describe two classification models (a Convolutional Neural Network a... more English. In this article, we describe two classification models (a Convolutional Neural Network and a Logistic Regression classifier), arranged according to three different strategies, submitted to subtask A of Automatic Misogyny Identification at EVALITA 2020. Results were very encouraging for detecting misogyny, even though aggressiveness was less accurate. Our second strategy, consisting of a Convolutional Neural Network and logistic regression to identify misogyny and aggressiveness, respectively, won the sixth place in the competition. Italiano. In questo articolo, descriviamo due modelli di classificazione (i.e., Convolutional Neural Network e Regressione Logistica), organizzati secondo tre diverse strategie, per il subtask A dello shared task Automatic Misogyny Identification a EVALITA 2020. I risultati sono stati molto incoraggianti nel rilevamento della misoginia, anche se l'aggressività viene riconosciuta con una precisione più basse. La nostra seconda strategia (Convolutional Neural Network per misoginia e Regressione Logistica per aggressività) ci ha permesso di ottenere il sesto posto nella competizione.
Anais do XXXII Simpósio Brasileiro de Informática na Educação (SBIE 2021), 2021
In this article, we present a systematic literature review, carried out from February to March 20... more In this article, we present a systematic literature review, carried out from February to March 2020, on the application of a machine learning technique to predict student dropout in higher education institutions. Besides describing the protocol followed during our research, which includes the research questions, searched databases and query strings, along with criteria for inclusion and exclusion of articles, we also present our main results, in terms of the attributes used by current research on this theme, along with adopted approaches, specific algorithms, and evalution metrics. The Decision Tree technique is the most used for the construction of models, and accuracy and recall and precision being the most used metric for evaluating models.
Anais do XIII Simpósio Brasileiro de Tecnologia da Informação e da Linguagem Humana (STIL 2021), 2021
A anotação de textos informais segundo a Universal Dependencies requer dois processos de segmenta... more A anotação de textos informais segundo a Universal Dependencies requer dois processos de segmentação: delimitação da unidade relevante para a análise sintática e identificação das palavras sintáticas. Neste artigo, apresentam-se as idiossincrasias linguísticas do corpus DANTEStocks, composto por tweets do mercado financeiro, escritos em Português, e as estratégias gerais de segmentação automática. Assim, contribui-se para a descrição de aspectos linguísticos dos tweets e para o desenvolvimento de recursos e ferramentas de processamento automático desse subgênero de "user-generated content".
Stock price forecasting is a non-linear, dynamic, complex, non-parametric and chaotic problem, be... more Stock price forecasting is a non-linear, dynamic, complex, non-parametric and chaotic problem, being one of the biggest challenges both in financial, academic, and technological terms, especially in what concerns decision support tools. In fact, only few investors are capable of delivering above-market performances in the long run. Within this context, this project aims at modelling a Multi-layer Perceptron Artificial Neural Network (MLP-ANN) capable of forecasting IBOVESPA’s closure trend, based on closure trends of a set of international indexes. Starting from a set of twelve indexes, from three continents, we modelled ANNs that take as input the closure trend of one index per continent, classifying IBOVESPA’s next closure as higher or lower than the closure in the day before. Along with different input sets, ANNs were tested with different training parameters. We then compare the results with those by (Malagrino, 2012), in which the same data set was used to train a Bayesian Netw...
DG.O (Inter)National Conference on Digital Government Research, 2009
With the growing importance of foreign commerce comes also greater opportunities for fraudulent b... more With the growing importance of foreign commerce comes also greater opportunities for fraudulent behaviour. As such, governments must try to detect frauds as soon as they take place, if they are to avoid the profound damage to the so- ciety frauds may cause. Although current fraud detection systems can be used on this endeavour with reasonable ac- curacy, they still
DG.O (Inter)National Conference on Digital Government Research, 2008
There is an increasing concern about the control of customs operations. While globalization incen... more There is an increasing concern about the control of customs operations. While globalization incentives the opening of the market, increasing amounts of imports and exports have been used to conceal several illicit activities, such as, tax evasion, smuggling, money laundry, and drug tra!c. This fact makes it paramount for governments to find automatic or semi-automatic solutions to guide the customs'
In this paper, we introduce a corpus of human-authored dialogue summaries collected through a web... more In this paper, we introduce a corpus of human-authored dialogue summaries collected through a web-experiment. The corpus features (i) one of the few existing corpora of written dialogue summaries; (ii) the only corpus available for dialogue summaries in Portuguese; and (iii) the only available corpus of summaries produced for dialogues whose participants' politeness alignment was systematically varied. Comprising 1,808 human-authored summaries, produced by 452 summarisers, for four different dialogues, this is, to the best of our knowledge, the largest individual corpus available for dialogue summaries, with the highest number of participants involved.
Uploads
Papers by Norton Roman