Neural Question Generation for the Portuguese Language: A Preliminary Study

Abstract

Question Generation aims to automatically generate questions based on a given input provided as context. A controllable question generation scheme focuses on generating questions with specific attributes, allowing better control. In this study, we propose a few-shot prompting strategy for controlling the generation of question-answer pairs from children's narrative texts. We aim to control two attributes: the question's explicitness and underlying narrative elements. With empirical evaluation, we show the effectiveness of controlling the generation process by employing few-shot prompting side by side with a reference model. Our experiments highlight instances where the few-shot strategy surpasses the reference model, particularly in scenarios such as semantic closeness evaluation and the diversity and coherency of question-answer pairs. However, these improvements are not always statistically significant. The code is publicly available at github.com/bernardoleite/few-shot-prompting-qg-control.

Feedback, Control, or Explanations? Supporting Teachers With Steerable Distractor-Generating AI

“…Lack of transparency is often mentioned as an additional barrier, as teachers have limited insight into, or control over, the decision-making process of such model. Recent work has put more effort into investigating ways to give teachers control over question and distractor generating models, allowing them to improve results and even increase creativity [22,26]. However, these have not focused on the human-computer interaction aspect, leaving many open research questions on how to enable effective user control and model understanding.…”

“…Second, interactive machine learning (IML) facilitates closer human-AI collaboration by involving users in the model training phase and allowing them to integrate their domain expertise to improve model performance [21]. Third, users can control model outcomes with filtering or sorting mechanisms that steer outcomes without retraining the underlying model [26].…”

Exploring BERT for Aspect Extraction in Portuguese Language

“…Moving towards larger encoderdecoder models, Carmo et al (2020) proposed the PTT5 model, based on the T5 architecture. It was then fine-tuned for paraphrasing tasks (Schneider et al, 2021), and for Portuguese questiongeneration (Leite and Lopes Cardoso, 2022).…”