A primary focus of machine reading comprehension (MRC) research centers around the retrieval of declarative knowledge - that is, explicitly stated or static descriptions of entities within a knowledge-base (KB). Evidence suggests that current MRC and neural methods struggle to fully engage in actual comprehension.
The majority of current MRC datasets effectively train models to perform simple pattern matching of words & phrases when attempting to query a knowledge source. This fails to mimic the manner in which humans use interaction and observation in order to gain knowledge about an environment. Furthermore, this simple pattern recognition does not reward information-seeking behaviour that is necessary to answer many natural language questions.
Interactive Question Answering (IQA) has been proposed as a viable solution to the lack of comprehension skills current methods fail to develop. IQA sees an agent be tasked with answering questions that require interacting with some dynamic environment such as text-based games. Text-based games have risen in popularity by allowing for language-learning problems to be approached using reinforcement learning (RL) methods in dialogue-like environments.
The QAit (Question Answering with interactive text) task proposes a novel text-based question answering problem whereby an agent must interact with a partially observable text-based environment to gather the declarative knowledge required to answer questions.
Learn moreThis study explores the effectiveness of policy-based methods within textual environments and the use of a predictive environment dynamics model in the creation of generalisable and language comprehending agents. Prior work has shown policy-based methods to have significantly better generalisation capabilities than their value-based counterparts. Additionally, the use of a predictive environment dynamics model as proposed by Pathak et al (2017). has been seen to further aid generalisation.
Learn moreThis study explores the use of Graph Attention Networks (GATs) in providing an RL agent with some contextual understanding about the environment with which it inhabits. Literature shows that GATs can aid aspects of performance such as steps required to complete a task and training convergence. Therefore, by embedding specific details about an environment into a knowledge graph and allowing an agent to use it to inform decision-making, this study explores the effects of Graph Attention Networks on agents’ accuracy on the QAit task.
Learn moreThis study proposes replacing the original QA module, in the QAit baseline architecture, with a fine-tuned BERT model, aiming to leverage pre-trained word embeddings and language understanding to provide more accurate answers to questions. We will replace the role of the RL agent with a Decision Transformer that utilises the GPT-2 architecture and will closely follow the methodology outlined by Chen et al (2021).
Learn more