Fabio G. Cozman

Paulo Pirozelli, Marcos M. José, Igor Silveira et al.

Pirá is a reading comprehension dataset focused on the ocean, the Brazilian coast, and climate change, built from a collection of scientific abstracts and reports on these topics. This dataset represents a versatile language resource, particularly useful for testing the ability of current machine learning models to acquire expert scientific knowledge. Despite its potential, a detailed set of baselines has not yet been developed for Pirá. By creating these baselines, researchers can more easily utilize Pirá as a resource for testing machine learning models across a wide range of question answering tasks. In this paper, we define six benchmarks over the Pirá dataset, covering closed generative question answering, machine reading comprehension, information retrieval, open question answering, answer triggering, and multiple choice question answering. As part of this effort, we have also produced a curated version of the original dataset, where we fixed a number of grammar issues, repetitions, and other shortcomings. Furthermore, the dataset has been extended in several new directions, so as to face the aforementioned benchmarks: translation of supporting texts from English into Portuguese, classification labels for answerability, automatic paraphrases of questions and answers, and multiple choice candidates. The results described in this paper provide several points of reference for researchers interested in exploring the challenges provided by the Pirá dataset.

Paulo Pirozelli, Ais B. R. Castro, Ana Luiza C. de Oliveira et al.

We describe the first steps in the development of an artificial agent focused on the Brazilian maritime territory, a large region within the South Atlantic also known as the Blue Amazon. The "BLue Amazon Brain" (BLAB) integrates a number of services aimed at disseminating information about this region and its importance, functioning as a tool for environmental awareness. The main service provided by BLAB is a conversational facility that deals with complex questions about the Blue Amazon, called BLAB-Chat; its central component is a controller that manages several task-oriented natural language processing modules (e.g., question answering and summarizer systems). These modules have access to an internal data lake as well as to third-party databases. A news reporter (BLAB-Reporter) and a purposely-developed wiki (BLAB-Wiki) are also part of the BLAB service architecture. In this paper, we describe our current version of BLAB's architecture (interface, backend, web services, NLP modules, and resources) and comment on the challenges we have faced so far, such as the lack of training data and the scattered state of domain information. Solving these issues presents a considerable challenge in the development of artificial intelligence for technical domains.

Marlon S. Mathias, Marcel R. de Barros, Jefferson F. Coelho et al.

We describe a Physics-Informed Neural Network (PINN) that simulates the flow induced by the astronomical tide in a synthetic port channel, with dimensions based on the Santos - São Vicente - Bertioga Estuarine System. PINN models aim to combine the knowledge of physical systems and data-driven machine learning models. This is done by training a neural network to minimize the residuals of the governing equations in sample points. In this work, our flow is governed by the Navier-Stokes equations with some approximations. There are two main novelties in this paper. First, we design our model to assume that the flow is periodic in time, which is not feasible in conventional simulation methods. Second, we evaluate the benefit of resampling the function evaluation points during training, which has a near zero computational cost and has been verified to improve the final model, especially for small batch sizes. Finally, we discuss some limitations of the approximations used in the Navier-Stokes equations regarding the modeling of turbulence and how it interacts with PINNs.

Felipe M. Moreno, Caio F. D. Netto, Marcel R. de Barros et al.

In this work we improve forecasting of Sea Surface Height (SSH) and current velocity (speed and direction) in oceanic scenarios. We do so by resorting to Random Forests so as to predict the error of a numerical forecasting system developed for the Santos Channel in Brazil. We have used the Santos Operational Forecasting System (SOFS) and data collected in situ between the years of 2019 and 2021. In previous studies we have applied similar methods for current velocity in the channel entrance, in this work we expand the application to improve the SHH forecast and include four other stations in the channel. We have obtained an average reduction of 11.9% in forecasting Root-Mean Square Error (RMSE) and 38.7% in bias with our approach. We also obtained an increase of Agreement (IOA) in 10 of the 14 combinations of forecasted variables and stations.

Marcos M. José, Flávio N. Cação, Maria F. Ribeiro et al.

This paper proposes a novel architecture to generate multi-hop answers to open domain questions that require information from texts and tables, using the Open Table-and-Text Question Answering dataset for validation and training. One of the most common ways to generate answers in this setting is to retrieve information sequentially, where a selected piece of data helps searching for the next piece. As different models can have distinct behaviors when called in this sequential information search, a challenge is how to select models at each step. Our architecture employs reinforcement learning to choose between different state-of-the-art tools sequentially until, in the end, a desired answer is generated. This system achieved an F1-score of 19.03, comparable to iterative systems in the literature.

Yan V. Sym, João Gabriel M. Campos, Fabio G. Cozman

This demo paper introduces the BLAB Reporter, a robot-journalist covering the Brazilian Blue Amazon. The Reporter is based on a pipeline architecture for Natural Language Generation; it offers daily reports, news summaries and curious facts in Brazilian Portuguese. By collecting, storing and analysing structured data from publicly available sources, the robot-journalist uses domain knowledge to generate and publish texts in Twitter. Code and corpus are publicly available

Yan V. Sym, João Gabriel M. Campos, Marcos M. José et al.

The majority of NLG systems have been designed following either a template-based or a pipeline-based architecture. Recent neural models for data-to-text generation have been proposed with an end-to-end deep learning flavor, which handles non-linguistic input in natural language without explicit intermediary representations. This study compares the most often employed methods for generating Brazilian Portuguese texts from structured data. Results suggest that explicit intermediate steps in the generation process produce better texts than the ones generated by neural end-to-end architectures, avoiding data hallucination while better generalizing to unseen inputs. Code and corpus are publicly available.

Paulo Pirozelli, Victor Hugo Nascimento Rocha, Fabio G. Cozman et al.

Arguments are a fundamental aspect of human reasoning, in which claims are supported, challenged, and weighed against one another. We present an end-to-end large language model (LLM)-based system for reconstructing arguments from natural language text into abstract argument graphs. The system follows a multi-stage pipeline that progressively identifies argumentative components, selects relevant elements, and uncovers their logical relations. These elements are represented as directed acyclic graphs consisting of two component types (premises and conclusions) and three relation types (support, attack, and undercut). We conduct two complementary experiments to evaluate the system. First, we perform a manual evaluation on arguments drawn from an argumentation theory textbook to assess the system's ability to recover argumentative structure. Second, we conduct a quantitative evaluation on benchmark datasets, allowing comparison with prior work by mapping our outputs to established annotation schemes. Results show that the system can adequately recover argumentative structures and, when adapted to different annotation schemes, achieve reasonable performance across benchmark datasets. These findings highlight the potential of LLM-based pipelines for scalable argument reconstruction.

Paulo Pirozelli, Marcos M. José, Paulo de Tarso P. Filho et al.

Logical reasoning is central to complex human activities, such as thinking, debating, and planning; it is also a central component of many AI systems as well. In this paper, we investigate the extent to which encoder-only transformer language models (LMs) can reason according to logical rules. We ask whether those LMs can deduce theorems in propositional calculus and first-order logic; if their relative success in these problems reflects general logical capabilities; and which layers contribute the most to the task. First, we show for several encoder-only LMs that they can be trained, to a reasonable degree, to determine logical validity on various datasets. Next, by cross-probing fine-tuned models on these datasets, we show that LMs have difficulty in transferring their putative logical reasoning ability, which suggests that they may have learned dataset-specific features, instead of a general capability. Finally, we conduct a layerwise probing experiment, which shows that the hypothesis classification task is mostly solved through higher layers.

Gustavo Kuratomi, Paulo Pirozelli, Fabio G. Cozman et al.

Although large language models (LLMs) demonstrate strong text generation capabilities, they struggle in scenarios requiring access to structured knowledge bases or specific documents, limiting their effectiveness in knowledge-intensive tasks. To address this limitation, retrieval-augmented generation (RAG) models have been developed, enabling generative models to incorporate relevant document fragments into their inputs. In this paper, we design and evaluate a RAG-based virtual assistant specifically tailored for the University of São Paulo. Our system architecture comprises two key modules: a retriever and a generative model. We experiment with different types of models for both components, adjusting hyperparameters such as chunk size and the number of retrieved documents. Our optimal retriever model achieves a Top-5 accuracy of 30%, while our most effective generative model scores 22.04\% against ground truth answers. Notably, when the correct document chunks are supplied to the LLMs, accuracy significantly improves to 54.02%, an increase of over 30 percentage points. Conversely, without contextual input, performance declines to 13.68%. These findings highlight the critical role of database access in enhancing LLM performance. They also reveal the limitations of current semantic search methods in accurately identifying relevant documents and underscore the ongoing challenges LLMs face in generating precise responses.

João P. Arroyo, João G. Rodrigues, Daniel Lawand et al.

We investigate partially identifiable queries in a class of causal models. We focus on acyclic Structural Causal Models that are quasi-Markovian (that is, each endogenous variable is connected with at most one exogenous confounder). We look into scenarios where endogenous variables are observed (and a distribution over them is known), while exogenous variables are not fully specified. This leads to a representation that is in essence a Bayesian network where the distribution of root variables is not uniquely determined. In such circumstances, it may not be possible to precisely compute a probability value of interest. We thus study the computation of tight probability bounds, a problem that has been solved by multilinear programming in general, and by linear programming when a single confounded component is intervened upon. We present a new algorithm to simplify the construction of such programs by exploiting input probabilities over endogenous variables. For scenarios with a single intervention, we apply column generation to compute a probability bound through a sequence of auxiliary linear integer programs, thus showing that a representation with polynomial cardinality for exogenous variables is possible. Experiments show column generation techniques to be superior to existing methods.

André F. A. Paschoal, Paulo Pirozelli, Valdinei Freire et al.

Current research in natural language processing is highly dependent on carefully produced corpora. Most existing resources focus on English; some resources focus on languages such as Chinese and French; few resources deal with more than one language. This paper presents the Pirá dataset, a large set of questions and answers about the ocean and the Brazilian coast both in Portuguese and English. Pirá is, to the best of our knowledge, the first QA dataset with supporting texts in Portuguese, and, perhaps more importantly, the first bilingual QA dataset that includes this language. The Pirá dataset consists of 2261 properly curated question/answer (QA) sets in both languages. The QA sets were manually created based on two corpora: abstracts related to the Brazilian coast and excerpts of United Nation reports about the ocean. The QA sets were validated in a peer-review process with the dataset contributors. We discuss some of the advantages as well as limitations of Pirá, as this new resource can support a set of tasks in NLP such as question-answering, information retrieval, and machine translation.

Juliana Cesaro, Fabio G. Cozman

One often finds in the literature connections between measures of fairness and measures of feature importance employed to interpret trained classifiers. However, there seems to be no study that compares fairness measures and feature importance measures. In this paper we propose ways to evaluate and compare such measures. We focus in particular on SHAP, a game-theoretic measure of feature importance; we present results for a number of unfairness-prone datasets.

Francisco H. O. V. de Faria, Arthur C. Gusmão, Fabio G. Cozman et al.

ProbLog is a state-of-art combination of logic programming and probabilities; in particular ProbLog offers parameter learning through a variant of the EM algorithm. However, the resulting learning algorithm is rather slow, even when the data are complete. In this short paper we offer some insights that lead to orders of magnitude improvements in ProbLog's parameter learning speed with complete data.