News media employ moral language to create memorable stories, and readers often engage with the content that align with their values. Moral theories have been applied to news analysis studying moral values in isolation, while the intricate dynamics among participating entities in shaping moral events have been overlooked. This is mainly due to the use of obscure language to conceal evident ideology and values, coupled with the insufficient moral reasoning capability in most existing NLP systems, where LLMs are no exception. To study this phenomenon, we first annotate a new dataset, MORAL EVENTS, consisting of 5,494 structured annotations on 474 news articles by diverse US media across the political spectrum. We further propose MOKA, a moral event extraction framework with MOral Knowledge Augmentation, that leverages knowledge derived from moral words and moral scenarios. Experimental results show that MOKA outperforms competitive baselines across three moral event understanding tasks. Further analyses illuminate the selective reporting of moral events by media outlets of different ideological leanings, suggesting the significance of event-level morality analysis in news. Our datasets and codebase are available at https://github.com/launchnlp/MOKA.

Pre-trained language models (PLMs) have established the new paradigm in the field of NLP. For more powerful PLMs, one of the most popular and successful way is to continuously scale up sizes of the models and the pre-training corpora. These large corpora are generally obtained by converging smaller ones from multiple sources, they are thus growing increasingly diverse. However, the side-effects of these colossal converged corpora remain understudied. In this paper, we identify the disadvantage of heterogeneous corpora from multiple sources for pre-training PLMs. Towards coordinated pre-training on diverse corpora, we further propose source prompts (SP), which explicitly prompt the model of the data source at the pre-training and fine-tuning stages. Results of extensive experiments demonstrate that PLMs pre-trained with SP on diverse corpora gain significant improvement in various downstream tasks.

Large Language Models (LLMs) often struggle when faced with situations where they lack the prerequisite knowledge to generate a sensical response. In these cases, models tend to fabricate and hallucinate, rather than appropriately signaling uncertainty as humans would. This behavior misaligns with human conversational norms and presents challenges surrounding responsible and ethical AI development. This work aims to systematically investigate LLMs' behaviors in such situations. We curate an adversarial question-answering benchmark containing unanswerable questions targeting information absent from the LLM's training data. Concretely, these unanswerable questions contain non-existent concepts or false premises. When presented with such unanswerable questions, an LLM should appropriately convey uncertainty, and be able to challenge the premise and refuse to generate a response. While facing answerable valid questions, a model should demonstrate a positive correlation between accuracy and confidence. Using a model-agnostic unified confidence elicitation approach, we observe that LLMs that have gone through instruction finetuning and reinforcement learning from human feedback (RLHF) perform significantly better than their counterparts that do not. Moreover, uncertainty expression 1 through our elicitation method does not always stay consistent with the perceived confidence of the direct response of an LLM. Our findings call for further research into teaching LLMs to proactively and reliably express uncertainty.

Human perception of language depends on personal backgrounds like gender and ethnicity. While existing studies have shown that large language models (LLMs) hold values that are closer to certain societal groups, it is unclear whether their prediction behaviors on subjective NLP tasks also exhibit a similar bias. In this study, leveraging the POPQUORN dataset which contains annotations of diverse demographic backgrounds, we conduct a series of experiments on four popular LLMs to investigate their capability to understand group differences and potential biases in their predictions for politeness and offensiveness. We find that for both tasks, model predictions are closer to the labels from White and female participants. We further explore prompting with the target demographic labels and show that including the target demographic in the prompt actually worsens the model's performance. More specifically, when being prompted to respond from the perspective of "Black" and "Asian" individuals, models show lower performance in predicting both overall scores as well as the scores from corresponding groups. Our results suggest that LLMs hold gender and racial biases for subjective NLP tasks and that demographic-infused prompts alone may be insufficient to mitigate such effects. Code and data are available at https://github.com/Jiaxin-Pei/LLM-Group-Bias.

Large language models (LLMs) often struggle with maintaining accuracy across a sequence of intermediate reasoning steps in mathematical reasoning, leading to error propagation that undermines the final result. The current methodology to mitigate this issue primarily involves using a verifier model to assess the correctness of generated solution candidates, focusing either on the overall reasoning path or on an incomplete reasoning path. By rethinking this approach, we argue that assessing potentials of incomplete reasoning paths could be more advantageous as it guides towards correct final answers, transforming the task into a \textit{planning} problem. Our proposed verifier, the Outcome-supervision Value Model (OVM), employs outcome supervision for training, offering an efficient and intuitive method for \textit{planning} by prioritizing steps that lead to accurate conclusions over mere per-step correctness. Furthermore, the OVM eschews the need for labor-intensive annotations on step-level correctness, enhancing its scalability. Our experiments on two multi-step mathematical reasoning datasets, GSM8K and Game of 24, demonstrate the superior performance of the OVM model. Notably, in GSM8K, our \textbf{OVM-7B model achieves state-of-the-art results among LLMs up to 13B parameters}; especially it does not utilize GPT-4 or code execution. These findings offer a novel perspective on the role of outcome supervision in training verifiers for multi-step reasoning tasks and provide theoretical justification for its advantage in value estimation for planning.

This study introduces a new long-form database question answering dataset designed to evaluate how Large Language Models (LLMs) interact with a SQL interpreter. The task necessitates LLMs to strategically generate multiple SQL queries to retrieve sufficient data from a database, to reason with the acquired context, and to synthesize them into a comprehensive analytical narrative. Our findings highlight that this task poses great challenges even for the state-of-the-art GPT-4 model. We propose and evaluate two interaction strategies, and provide a fine-grained analysis of the individual stages within the interaction. A key discovery is the identification of two primary bottlenecks hindering effective interaction: the capacity for planning and the ability to generate multiple SQL queries. To address the challenge of accurately assessing answer quality, we introduce a multi-agent evaluation framework that simulates the academic peer-review process, enhancing the precision and reliability of our evaluations. This framework allows for a more nuanced understanding of the strengths and limitations of current LLMs in complex retrieval and reasoning tasks.

Language models can serve as a valuable tool for software developers to increase productivity. Large generative models can be used for code generation and code completion, while smaller encoder-only models are capable of performing code search tasks using natural language queries.These capabilities are heavily influenced by the quality and diversity of the available training data. Source code datasets used for training usually focus on the most popular languages and testing is mostly conducted on the same distributions, often overlooking low-resource programming languages. Motivated by the NLP generalization taxonomy proposed by Hupkes et.\,al., we propose a new benchmark dataset called GenCodeSearchNet (GeCS) which builds upon existing natural language code search datasets to systemically evaluate the programming language understanding generalization capabilities of language models. As part of the full dataset, we introduce a new, manually curated subset StatCodeSearch that focuses on R, a popular but so far underrepresented programming language that is often used by researchers outside the field of computer science. For evaluation and comparison, we collect several baseline results using fine-tuned BERT-style models and GPT-style large language models in a zero-shot setting.

Despite the recent advancement in large language models (LLMs) and their high performances across numerous benchmarks, recent research has unveiled that LLMs suffer from hallucinations and unfaithful reasoning. This work studies a specific type of hallucination induced by semantic associations. Specifically, we investigate to what extent LLMs take shortcuts from certain keyword/entity biases in the prompt instead of following the correct reasoning path. To quantify this phenomenon, we propose a novel probing method and benchmark called EureQA. We start from questions that LLMs will answer correctly with utmost certainty, and mask the important entity with evidence sentence recursively, asking models to find masked entities according to a chain of evidence before answering the question. During the construction of the evidence, we purposefully replace semantic clues (entities) that may lead to the correct answer with distractor clues (evidence) that will not directly lead to the correct answer but require a chain-like reasoning process. We evaluate if models can follow the correct reasoning chain instead of short-cutting through distractor clues. We find that existing LLMs lack the necessary capabilities to follow correct reasoning paths and resist the attempt of greedy shortcuts. We show that the distractor semantic associations often lead to model hallucination, which is strong evidence that questions the validity of current LLM reasoning.

Are the longstanding robustness issues in NLP resolved by today's larger and more performant models? To address this question, we conduct a thorough investigation using 19 models of different sizes spanning different architectural choices and pretraining objectives. We conduct evaluations using (a) OOD and challenge test sets, (b) CheckLists, (c) contrast sets, and (d) adversarial inputs. Our analysis reveals that not all OOD tests provide further insight into robustness. Evaluating with CheckLists and contrast sets shows significant gaps in model performance; merely scaling models does not make them sufficiently robust. Finally, we point out that current approaches for adversarial evaluations of models are themselves problematic: they can be easily thwarted, and in their current forms, do not represent a sufficiently deep probe of model robustness. We conclude that not only is the question of robustness in NLP as yet unresolved, but even some of the approaches to measure robustness need to be reassessed.

We find that the best publicly available LLMs like GPT-4 and PaLM 2 currently perform poorly at basic text handling required of lawyers or paralegals, such as looking up the text at a line of a witness deposition or at a subsection of a contract. We introduce a benchmark to quantify this poor performance, which casts into doubt LLMs' current reliability as-is for legal practice. Finetuning for these tasks brings an older LLM to near-perfect performance on our test set and also raises performance on a related legal task. This stark result highlights the need for more domain expertise in LLM training.