Although moral responsibility is not circumscribed by causality, they are both closely intermixed. Furthermore, rationally understanding the evolution of the physical world is inherently linked with the idea of causality. Thus, the decision-making applications based on automated planning inevitably have to deal with causality, especially if they consider imputability aspects or integrate references to ethical norms. The many debates around causation in the last decades have shown how complex this notion is and thus, how difficult is its integration with planning. As a result, much of the work in computational ethics relegates causality to the background, despite the considerations stated above. This paper's contribution is to provide a complete and sound translation into logic programming from an actual causation definition suitable for action languages, this definition is a formalisation of Wright's NESS test. The obtained logic program allows to deal with complex causal relations. In addition to enabling agents to reason about causality, this contribution specifically enables the computational ethics domain to handle situations that were previously out of reach. In a context where ethical considerations in decision-making are increasingly important, advances in computational ethics can greatly benefit the entire AI community.

Building Natural Language Understanding (NLU) capabilities for Indic languages, which have a collective speaker base of more than one billion speakers is absolutely crucial. In this work, we aim to improve the NLU capabilities of Indic languages by making contributions along 3 important axes (i) monolingual corpora (ii) NLU testsets (iii) multilingual LLMs focusing on Indic languages. Specifically, we curate the largest monolingual corpora, IndicCorp, with 20.9B tokens covering 24 languages from 4 language families - a 2.3x increase over prior work, while supporting 12 additional languages. Next, we create a human-supervised benchmark, IndicXTREME, consisting of nine diverse NLU tasks covering 20 languages. Across languages and tasks, IndicXTREME contains a total of 105 evaluation sets, of which 52 are new contributions to the literature. To the best of our knowledge, this is the first effort towards creating a standard benchmark for Indic languages that aims to test the multilingual zero-shot capabilities of pretrained language models. Finally, we train IndicBERT v2, a state-of-the-art model supporting all the languages. Averaged across languages and tasks, the model achieves an absolute improvement of 2 points over a strong baseline. The data and models are available at https://github.com/AI4Bharat/IndicBERT.

Privacy policies are long, complex documents that end-users seldom read. Privacy labels aim to ameliorate these issues by providing succinct summaries of salient data practices. In December 2020, Apple began requiring that app developers submit privacy labels describing their apps' data practices. Yet, research suggests that app developers often struggle to do so. In this paper, we automatically identify possible discrepancies between mobile app privacy policies and their privacy labels. Such discrepancies could be indicators of potential privacy compliance issues. We introduce the Automated Privacy Label Analysis System (ATLAS). ATLAS includes three components: a pipeline to systematically retrieve iOS App Store listings and privacy policies; an ensemble-based classifier capable of predicting privacy labels from the text of privacy policies with 91.3% accuracy using state-of-the-art NLP techniques; and a discrepancy analysis mechanism that enables a large-scale privacy analysis of the iOS App Store. Our system has enabled us to analyze 354,725 iOS apps. We find several interesting trends. For example, only 40.3% of apps in the App Store provide easily accessible privacy policies, and only 29.6% of apps provide both accessible privacy policies and privacy labels. Among apps that provide both, 88.0% have at least one possible discrepancy between the text of their privacy policy and their privacy label, which could be indicative of a potential compliance issue. We find that, on average, apps have 5.32 such potential compliance issues. We hope that ATLAS will help app developers, researchers, regulators, and mobile app stores alike. For example, app developers could use our classifier to check for discrepancies between their privacy policies and privacy labels, and regulators could use our system to help review apps at scale for potential compliance issues.

A large portion of society united around the same vision and ideas carries enormous energy. That is precisely what political figures would like to accumulate for their cause. With this goal in mind, they can sometimes resort to distorting or hiding the truth, unintentionally or on purpose, which opens the door for misinformation and disinformation. Tools for automatic detection of check-worthy claims would be of great help to moderators of debates, journalists, and fact-checking organizations. While previous work on detecting check-worthy claims has focused on text, here we explore the utility of the audio signal as an additional information source. We create a new multimodal dataset (text and audio in English) containing 48 hours of speech. Our evaluation results show that the audio modality together with text yields improvements over text alone in the case of multiple speakers. Moreover, an audio-only model could outperform a text-only one for a single speaker.

Acquiring high-quality data for training discriminative models is a crucial yet challenging aspect of building effective predictive systems. In this paper, we present Diffusion Inversion, a simple yet effective method that leverages the pre-trained generative model, Stable Diffusion, to generate diverse, high-quality training data for image classification. Our approach captures the original data distribution and ensures data coverage by inverting images to the latent space of Stable Diffusion, and generates diverse novel training images by conditioning the generative model on noisy versions of these vectors. We identify three key components that allow our generated images to successfully supplant the original dataset, leading to a 2-3x enhancement in sample complexity and a 6.5x decrease in sampling time. Moreover, our approach consistently outperforms generic prompt-based steering methods and KNN retrieval baseline across a wide range of datasets. Additionally, we demonstrate the compatibility of our approach with widely-used data augmentation techniques, as well as the reliability of the generated data in supporting various neural architectures and enhancing few-shot learning.

Large language models (LLMs) have made significant progress in natural language processing (NLP), and are utilized extensively in various applications. Recent works, such as chain-of-thought (CoT), have shown that intermediate reasoning steps can improve the performance of LLMs for complex reasoning tasks, such as math problems and symbolic question-answering tasks. However, we notice the challenge that LLMs face when it comes to temporal reasoning. Our preliminary experiments show that generating intermediate reasoning steps does not always boost the performance of complex temporal question-answering tasks. Therefore, we propose a novel framework that combines the extraction capability of LLMs and the logical reasoning capability of a Python solver to tackle this issue. Extensive experiments and analysis demonstrate the effectiveness of our framework in handling intricate time-bound reasoning tasks.

Sentiment analysis (SA) has been a long-standing research area in natural language processing. It can offer rich insights into human sentiments and opinions and has thus seen considerable interest from both academia and industry. With the advent of large language models (LLMs) such as ChatGPT, there is a great potential for their employment on SA problems. However, the extent to which existing LLMs can be leveraged for different sentiment analysis tasks remains unclear. This paper aims to provide a comprehensive investigation into the capabilities of LLMs in performing various sentiment analysis tasks, from conventional sentiment classification to aspect-based sentiment analysis and multifaceted analysis of subjective texts. We evaluate performance across 13 tasks on 26 datasets and compare the results against small language models (SLMs) trained on domain-specific datasets. Our study reveals that while LLMs demonstrate satisfactory performance in simpler tasks, they lag behind in more complex tasks requiring deeper understanding or structured sentiment information. However, LLMs significantly outperform SLMs in few-shot learning settings, suggesting their potential when annotation resources are limited. We also highlight the limitations of current evaluation practices in assessing LLMs' SA abilities and propose a novel benchmark, \textsc{SentiEval}, for a more comprehensive and realistic evaluation. Data and code during our investigations are available at \url{https://github.com/DAMO-NLP-SG/LLM-Sentiment}.

Past work has studied event prediction and event language modeling, sometimes mediated through structured representations of knowledge in the form of event schemas. Such schemas can lead to explainable predictions and forecasting of unseen events given incomplete information. In this work, we look at the process of creating such schemas to describe complex events. We use large language models (LLMs) to draft schemas directly in natural language, which can be further refined by human curators as necessary. Our focus is on whether we can achieve sufficient diversity and recall of key events and whether we can produce the schemas in a sufficiently descriptive style. We show that large language models are able to achieve moderate recall against schemas taken from two different datasets, with even better results when multiple prompts and multiple samples are combined. Moreover, we show that textual entailment methods can be used for both matching schemas to instances of events as well as evaluating overlap between gold and predicted schemas. Our method paves the way for easier distillation of event knowledge from large language model into schemas.

Tabular data is the most common format to publish and exchange structured data online. A clear example is the growing number of open data portals published by all types of public administrations. However, exploitation of these data sources is currently limited to technical people able to programmatically manipulate and digest such data. As an alternative, we propose the use of chatbots to offer a conversational interface to facilitate the exploration of tabular data sources. With our approach, any regular citizen can benefit and leverage them. Moreover, our chatbots are not manually created: instead, they are automatically generated from the data source itself thanks to the instantiation of a configurable collection of conversation patterns.

Obtaining labeled data to train a model for a task of interest is often expensive. Prior work shows training models on multitask data augmented with task descriptions (prompts) effectively transfers knowledge to new tasks. Towards efficiently building task-specific models, we assume access to a small number (32-1000) of unlabeled target-task examples and use those to retrieve the most similar labeled examples from a large pool of multitask data augmented with prompts. Compared to the current practice of finetuning models on uniformly sampled prompted multitask data (e.g.: FLAN, T0), our approach of finetuning on cross-task nearest neighbors is significantly more data-efficient. Using only 2% of the data from the P3 pool without any labeled target-task data, our models outperform strong baselines trained on all available data by 3-30% on 12 out of 14 datasets representing held-out tasks including legal and scientific document QA. Similarly, models trained on cross-task nearest neighbors from SuperNaturalInstructions, representing about 5% of the pool, obtain comparable performance to state-of-the-art models on 12 held-out tasks from that pool. Moreover, the models produced by our approach also provide a better initialization than single multitask finetuned models for few-shot finetuning on target-task data, as shown by a 2-23% relative improvement over few-shot finetuned T0-3B models on 8 datasets.