Long-Form Question Answering (LFQA) represents a growing interest in Legal Natural Language Processing (Legal-NLP) as many individuals encounter legal disputes at some point in their lives, but lack of knowledge about how to negotiate these complex situations might put them at risk. The endeavor to generate detailed answers to contextually rich legal questions has faced challenges, primarily due to the limited availability of specialized datasets involving intensive manual effort or incapability of existing LFQA models to produce informative responses. Addressing this, our research introduces a semi-synthetic dataset, Legal-LFQA (L2FQA) created by exploiting a large language model (LLM) and utilizing contexts derived from existing legal datasets. Additionally, we hypothesize that integrating legal reasoning into the answer generation process of the LLMs will help bolster both the quality and interpretability of the produced responses. We systematically analyze the quality of L2FQA using human evaluation and natural language inference based metrics. Next, we benchmark L2FQA on a wide range of general-purpose and domain-specific LLMs using fine-tuning and in-context learning (with zero, one and few shot) strategies. The efficacy of these techniques is gauged through several automated and human evaluations. Results indicate that incorporating legal reasoning into the answer generation process provides an avenue for improving the quality of responses in the context of Legal-LFQA task. By addressing the challenges faced in LFQA and emphasizing the potential of interpretability, this research contributes to the foundational work in enhancing question-answering systems within the legal domain.

Data-driven (deep) learning methods has led to parameterised abstractions of the data, often leading to stereotype societal biases in their predictions, e.g., predicting more frequently that women are weaker than men, or that African Americans are more likely to commit crimes than Caucasians. Standard approaches of mitigating such stereotypical biases from deep neural models include modifying the training dataset (pre-processing), or adjusting the model parameters with a bias-specific objective (in-processing). In our work, we approach this bias mitigation from a different perspective - that of an active learning-based selection of a subset of data instances towards training a model optimised for both effectiveness and fairness. Specifically speaking, the imbalances in the attribute value priors can be alleviated by constructing a balanced subset of the data instances with two selection objectives - first, of improving the model confidence of the primary task itself (a standard practice in active learning), and the second, of taking into account the parity of the model predictions with respect to the sensitive attributes, such as gender and race etc. We demonstrate that our proposed selection function achieves better results in terms of both the primary task effectiveness and fairness. The results are further shown to improve when this active learning-based data selection is combined with an in-process method of multi-objective training.

Large language models (LLMs), when scaled from millions to billions of parameters, have been demonstrated to exhibit the so-called 'emergence' effect, in that they are not only able to produce semantically correct and coherent text, but are also able to adapt themselves surprisingly well with small changes in contexts supplied as inputs (commonly called prompts). Despite producing semantically coherent and potentially relevant text for a given context, LLMs are vulnerable to yield incorrect information. This misinformation generation, or the so-called hallucination problem of an LLM, gets worse when an adversary manipulates the prompts to their own advantage, e.g., generating false propaganda to disrupt communal harmony, generating false information to trap consumers with target consumables etc. Not only does the consumption of an LLM-generated hallucinated content by humans pose societal threats, such misinformation, when used as prompts, may lead to detrimental effects for in-context learning (also known as few-shot prompt learning). With reference to the above-mentioned problems of LLM usage, we argue that it is necessary to foster research on topics related to not only identifying misinformation from LLM-generated content, but also to mitigate the propagation effects of this generated misinformation on downstream predictive tasks thus leading to more robust and effective leveraging in-context learning.

The scarcity of Mental Health Professionals (MHPs) available to assist patients underlines the need for developing automated systems to help MHPs combat the grievous mental illness called Major Depressive Disorder. In this paper, we develop a Virtual Assistant (VA) that serves as a first point of contact for users who are depressed or disheartened. In support based conversations, two primary components have been identified to produce positive outcomes,empathy andmotivation. While empathy necessitates acknowledging the feelings of the users with a desire to help, imparting hope and motivation uplifts the spirit of support seekers in distress. A combination of these aspects will ensure generalized positive outcome and beneficial alliance in mental health support. The VA, thus, should be capable of generating empathetic and motivational responses, continuously demonstrating positive sentiment by the VA. The end-to-end system employs two mechanisms in a pipe-lined manner : (i)Motivational Response Generator (MRG) : a sentiment driven Reinforcement Learning (RL) based motivational response generator; and (ii)Empathetic Rewriting Framework (ERF) : a transformer based model that rewrites the response from MRG to induce empathy. Experimental results indicate that our proposed VA outperforms several of its counterparts. To the best of our knowledge, this is the first work that seeks to incorporate these aspects together in an end-to-end system.