Recently, several public datasets for automatic speech recognition (ASR) in Brazilian Portuguese (BP) have been released, improving ASR systems performance. However, these datasets lack diversity in terms of age groups, regional accents, and education levels. In this paper, we present a new publicly available dataset consisting of 289 life story interviews (365 hours), featuring a broad range of speakers varying in age, education, and regional accents. First, we demonstrated the presence of bias in current BP ASR models concerning education levels and age groups. Second, we showed that our dataset helps mitigate these biases. Additionally, an ASR model trained on our dataset performed better during evaluation on a diverse test set. Finally, the ASR model trained with our dataset was extrinsically evaluated through a topic modeling task that utilized the automatically transcribed output.

Context: For thousands of years humans have been using natural language to register their knowledge on important information to enable its access to future generations. With internet, a large amount of textual data is produced and shared on a daily basis. So, scientists started to research techniques for efficiently process knowledge stored in textual format. In this context, Natural Language Processing (NLP) became a popular area studying linguistic phenomena and using computational methods to process texts in natural language. In particular, Open Information Extraction (Open IE) was proposed to gather information from plain text. Despite the advances in this area, it is still necessary to map details about how these approaches were proposed to support the community while creating more efficient Open IE systems. Objective: In this paper, we identify, in the literature, the main characteristics of proposed Open IE approaches. Method: First, we extended the search performed in a systematic mapping previously published by using backward snowballing and a manual search. Next, we updated the electronic database search including ACL Anthology. Finally, 159 studies proposing Open IE approaches were considered for data extraction. Results: Data analysis showed a significant increase in the number of studies published about Open IE in the last years. In addition, we provide important details about how these techniques were proposed (e.g., data sets used and output evaluation techniques). Results indicate that researchers started to adopt neural networks to perform Open IE instead of using conventional supervised learning techniques. Conclusion: Recent advances in Artificial Intelligence and neural networks techniques allowed scientists to have a new perspective on how to perform efficient textual data management. Therefore, Open IE approaches gained much attention as they can help in many contexts, especially in knowledge management tasks.

Context: Secondary studies, as Systematic Literature Reviews (SLRs) and Systematic Mappings (SMs), have been providing methodological and structured processes to identify and select research evidence in Computer Science, especially in Software Engineering (SE). One of the main activities of a secondary study process is to read the abstracts to decide on including or excluding studies. This activity is considered costly and time-consuming. In order to speed up the selection activity, some alternatives such as, structured abstracts and graphical abstracts (e.g. Concept Maps – CMs), have been proposed. Objective: This study presents an approach to automatically build CMs using Natural Language Processing (NLP) to support the selection activity of secondary studies. Method: First, we proposed an approach composed by two pipelines: (1) perform the triple extraction of concept-relation-concept based on NLP; and (2) attach the extracted triples in a structure used as a template to scientific studies. Second, we evaluated both pipelines conducting experiments. Results: The preliminary evaluation revealed that CMs extracted are coherent when compared with their source text. Conclusions: NLP can assist the automatic construction of CMs. In addition, the experiment results show that the approach can be useful to support researchers in the selection of studies in the selection activity of secondary studies.

In this paper, we propose SC-GlowTTS: an efficient zero-shot multi-speaker text-to-speech model that improves similarity for speakers unseen during training. We propose a speaker-conditional architecture that explores a flow-based decoder that works in a zero-shot scenario. As text encoders, we explore a dilated residual convolutional-based encoder, gated convolutional-based encoder, and transformer-based encoder. Additionally, we have shown that adjusting a GAN-based vocoder for the spectrograms predicted by the TTS model on the training dataset can significantly improve the similarity and speech quality for new speakers. Our model converges using only 11 speakers, reaching state-of-the-art results for similarity with new speakers, as well as high speech quality.

Sidney Evaldo Leal, João Marcos Munguba Vieira, Erica dos Santos Rodrigues, Elisângela Nogueira Teixeira, Sandra Aluísio. Proceedings of the 28th International Conference on Computational Linguistics. 2020.

Automatic analysis of connected speech by natural language processing techniques is a promising direction for diagnosing cognitive impairments. However, some difficulties still remain: the time required for manual narrative transcription and the decision on how transcripts should be divided into sentences for successful application of parsers used in metrics, such as Idea Density, to analyze the transcripts. The main goal of this paper was to develop a generic segmentation system for narratives of neuropsychological language tests. We explored the performance of our previous single-dataset-trained sentence segmentation architecture in a richer scenario involving three new datasets used to diagnose cognitive impairments, comprising different stories and two types of stimulus presentation for eliciting narratives — visual and oral — via illustrated story-book and sequence of scenes, and by retelling. Also, we proposed and evaluated three modifications to our previous RCNN architecture: (i) the inclusion of a Linear Chain CRF; (ii) the inclusion of a self-attention mechanism; and (iii) the replacement of the LSTM recurrent layer by a Quasi-Recurrent Neural Network layer. Our study allowed us to develop two new models for segmenting impaired speech transcriptions, along with an ideal combination of datasets and specific groups of narratives to be used as the training set.

Effective textual communication depends on readers being proficient enough to comprehend texts, and texts being clear enough to be understood by the intended audience, in a reading task. When the meaning of textual information and instructions is not well conveyed, many losses and damages may occur. Among the solutions to alleviate this problem is the automatic evaluation of sentence readability, task which has been receiving a lot of attention due to its large applicability. However, a shortage of resources, such as corpora for training and evaluation, hinders the full development of this task. In this paper, we generate a nontrivial sentence corpus in Portuguese. We evaluate three scenarios for building it, taking advantage of a parallel corpus of simplification, in which each sentence triplet is aligned and has simplification operations annotated, being ideal for justifying possible mistakes of future methods. The best scenario of our corpus PorSimplesSent is composed of 4,888 pairs, which is bigger than a similar corpus for English; all the three versions of it are publicly available. We created four baselines for PorSimplesSent and made available a pairwise ranking method, using 17 linguistic and psycholinguistic features, which correctly identifies the ranking of sentence pairs with an accuracy of 74.2%.

Leandro Santos, Edilson Anselmo Corrêa Júnior, Osvaldo Oliveira Jr, Diego Amancio, Letícia Mansur, Sandra Aluísio. Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2017.

Automated discourse analysis tools based on Natural Language Processing (NLP) aiming at the diagnosis of languageimpairing dementias generally extract several textual metrics of narrative transcripts.However, the absence of sentence boundary segmentation in the transcripts prevents the direct application of NLP methods which rely on these marks to function properly, such as taggers and parsers.We present the first steps taken towards automatic neuropsychological evaluation based on narrative discourse analysis, presenting a new automatic sentence segmentation method for impaired speech.Our model uses recurrent convolutional neural networks with prosodic, Part of Speech (PoS) features, and word embeddings.It was evaluated intrinsically on impaired, spontaneous speech, as well as, normal, prepared speech, and presents better results for healthy elderly (CTL) (F 1 = 0.74) and Mild Cognitive Impairment (MCI) patients (F 1 = 0.70) than the Conditional Random Fields method (F 1 = 0.55 and 0.53, respectively) used in the same context of our study.The results suggest that our model is robust for impaired speech and can be used in automated discourse analysis tools to differentiate narratives produced by MCI and CTL.

Idea Density (ID) was originally proposed as a way of measuring the memory load of narratives, by representing the underlying content of the text as a series of semantic units, called propositions or ideas. From a clinical perspective, this notion has been shown to correlate with several cognitive aspects, such as memory, readability, aging, and dementia onset and progress. Traditionally, propositions are extracted manually from texts. There is a tool that can automate ID extraction [1], but it uses shallow information as input, and doesn't produce the propositions themselves as output. We propose a novel approach to obtaining the ID automatically from a text. Our method is an automation of Chand et al.'s ID manual [2], and consists of a rule-based system acting upon dependency trees. Initially, for each sentence in a text, a dependency parser is used to elicit the dependency relations between words. Then, a set of rules is recursively applied in order to process these relations to yield the corresponding propositions. We analyze preliminary results of our system using a well-formed journalistic text, and speech transcriptions of dementia patients.

This paper describes the method employed to build a machinereadable pronunciation dictionary for Brazilian Portuguese. The dictionary makes use of a hybrid approach for converting graphemes into phonemes, based on both manual transcription rules and machine learning algorithms. It makes use of a word list compiled from the Portuguese Wikipedia dump. Wikipedia articles were transformed into plain text, tokenized and word types were extracted. A language identification tool was developed to detect loanwords among data. Words’ syllable boundaries and stress were identified. The transcription task was carried \nout in a two-step process: i) words are submitted to a set of transcription rules, in which predictable graphemes (mostly consonants) are transcribed; ii) a machine learning classifier is used to predict the transcription of the remaining graphemes (mostly vowels). The method was evaluated through 5-fold cross-validation; results show a F1-score of 0.98. The dictionary and all the resources used to build it were made publicly available.

In this paper we investigate the main linguistic phenomena that can make texts complex and how they could be simplified. We focus on a corpus analysis of simple account texts available on the web for Brazilian Portuguese and propose simplification strategies for this language. This study illustrates the need for text simplification to facilitate accessibility to information by poor literacy readers and potentially by people with other cognitive disabilities. It also highlights characteristics of simplification for Portuguese, which may differ from other languages. Such study consists of the first step towards building Brazilian Portuguese text simplification systems. One of the scenarios in which these systems could be used is that of reading electronic texts produced, e.g., by the Brazilian government or by relevant news agencies.

We present the computational and composition theoretical bases for the design of a collaborative writing tool, based on the critiquing approach, to assist non-native novice researchers in understanding and production of the structure of scientific papers. This critiquing tool is embedded in a suite named AMADEUS that caters for various needs of non-native English users to produce a first draft of a paper, relying on the reuse of contextualized linguistic material as input for the user. Our emphasis is on the architecture and methodology to build the linguistic resources for the critiquing tool. Though originally targeted at non-native authors, the critiquing tool may also be useful for novice native English writers and as a teaching resource for English for Academic Purposes practitioners.

The article presents the main features of a new environment for case-based reasoning systems development, RaBeCa. The software developed to run under Windows, aims to have an architecture open enough to accept external contributions and comprehend tests allowing integration with other artificial intelligence approaches. The system already uses a neural network. A small introduction to the case-based reasoning paradigm and a brief description of existing analogous systems are shown.