Identifying linguistic differences between dialects of a language often requires expert knowledge and meticulous human analysis. This is largely due to the complexity and nuance involved in studying various dialects. We present a novel approach to extract distinguishing lexical features of dialects by utilizing interpretable dialect classifiers, even in the absence of human experts. We explore both post-hoc and intrinsic approaches to interpretability, conduct experiments on Mandarin, Italian, and Low Saxon, and experimentally demonstrate that our method successfully identifies key language-specific lexical features that contribute to dialectal variations.

Large Language Models (LLMs) exhibit impressive capabilities but also present risks such as biased content generation and privacy issues. One of the current alignment techniques includes principle-driven integration, but it faces challenges arising from the imprecision of manually crafted rules and inadequate risk perception in models without safety training. To address these, we introduce Guide-Align, a two-stage approach. Initially, a safety-trained model identifies potential risks and formulates specific guidelines for various inputs, establishing a comprehensive library of guidelines and a model for input-guidelines retrieval. Subsequently, the retrieval model correlates new inputs with relevant guidelines, which guide LLMs in response generation to ensure safe and high-quality outputs, thereby aligning with human values. An additional optional stage involves fine-tuning a model with well-aligned datasets generated through the process implemented in the second stage. Our method customizes guidelines to accommodate diverse inputs, thereby enhancing the fine-grainedness and comprehensiveness of the guideline library. Furthermore, it incorporates safety expertise from a safety-trained LLM through a lightweight retrieval model. We evaluate our approach on three benchmarks, demonstrating significant improvements in LLM security and quality. Notably, our fine-tuned model, Labrador, even at 13 billion parameters, outperforms GPT-3.5-turbo and surpasses GPT-4 in alignment capabilities.

Logical reasoning has been an ongoing pursuit in the field of AI. Despite significant advancements made by large language models (LLMs), they still struggle with complex logical reasoning problems. To enhance reasoning performance, one promising direction is scalable oversight, which requires LLMs to identify their own errors and then improve by themselves. Various self-verification methods have been proposed in pursuit of this goal. Nevertheless, whether existing models understand their own errors well is still under investigation. In this paper, we take a closer look at the self-verification abilities of LLMs in the context of logical reasoning, focusing on their ability to identify logical fallacies accurately. We introduce a dataset, FALLACIES, containing 232 types of reasoning fallacies categorized in a hierarchical taxonomy. By conducting exhaustive experiments on FALLACIES, we obtain comprehensive and detailed analyses of a series of models on their verification abilities. Our main findings suggest that existing LLMs could struggle to identify fallacious reasoning steps accurately and may fall short of guaranteeing the validity of self-verification methods. Drawing from these observations, we offer suggestions for future research and practical applications of self-verification methods.

The field of imbalanced self-supervised learning, especially in the context of tabular data, has not been extensively studied. Existing research has predominantly focused on image datasets. This paper aims to fill this gap by examining the specific challenges posed by data imbalance in self-supervised learning in the domain of tabular data, with a primary focus on autoencoders. Autoencoders are widely employed for learning and constructing a new representation of a dataset, particularly for dimensionality reduction. They are also often used for generative model learning, as seen in variational autoencoders. When dealing with mixed tabular data, qualitative variables are often encoded using a one-hot encoder with a standard loss function (MSE or Cross Entropy). In this paper, we analyze the drawbacks of this approach, especially when categorical variables are imbalanced. We propose a novel metric to balance learning: a Multi-Supervised Balanced MSE. This approach reduces the reconstruction error by balancing the influence of variables. Finally, we empirically demonstrate that this new metric, compared to the standard MSE: i) outperforms when the dataset is imbalanced, especially when the learning process is insufficient, and ii) provides similar results in the opposite case.

Causal representation learning seeks to uncover latent, high-level causal representations from low-level observed data. It is particularly good at predictions under unseen distribution shifts, because these shifts can generally be interpreted as consequences of interventions. Hence leveraging {seen} distribution shifts becomes a natural strategy to help identifying causal representations, which in turn benefits predictions where distributions are previously {unseen}. Determining the types (or conditions) of such distribution shifts that do contribute to the identifiability of causal representations is critical. This work establishes a {sufficient} and {necessary} condition characterizing the types of distribution shifts for identifiability in the context of latent additive noise models. Furthermore, we present partial identifiability results when only a portion of distribution shifts meets the condition. In addition, we extend our findings to latent post-nonlinear causal models. We translate our findings into a practical algorithm, allowing for the acquisition of reliable latent causal representations. Our algorithm, guided by our underlying theory, has demonstrated outstanding performance across a diverse range of synthetic and real-world datasets. The empirical observations align closely with the theoretical findings, affirming the robustness and effectiveness of our approach.

Historians have started using machine learning to examine historical documents, including astronomical tables like those produced in Venice and other early modern cities. Proponents claim that the application of modern computer science to the past helps draw connections across a broader swath of the historical record than would otherwise be possible, correcting distortions that come from analyzing history one document at a time. But it introduces distortions of its own, including the risk that machine learning will slip bias or outright falsifications into the historical record. The way sea sponges pump water is really quite amazing. I never thought I'd be transfixed by a bed making competition, but here we are.

The rapid advancement of artificial intelligence (AI) and the expanding integration of large language models (LLMs) have ignited a debate about their application in education. This study delves into university instructors' experiences and attitudes toward AI language models, filling a gap in the literature by analyzing educators' perspectives on AI's role in the classroom and its potential impacts on teaching and learning. The objective of this research is to investigate the level of awareness, overall sentiment towardsadoption, and the factors influencing these attitudes for LLMs and generative AI-based tools in higher education. Data was collected through a survey using a Likert scale, which was complemented by follow-up interviews to gain a more nuanced understanding of the instructors' viewpoints. The collected data was processed using statistical and thematic analysis techniques. Our findings reveal that educators are increasingly aware of and generally positive towards these tools. We find no correlation between teaching style and attitude toward generative AI. Finally, while CS educators show far more confidence in their technical understanding of generative AI tools and more positivity towards them than educators in other fields, they show no more confidence in their ability to detect AI-generated work.

Large Language Models (LLMs) have achieved impressive performance across various reasoning tasks. However, even state-of-the-art LLMs such as ChatGPT are prone to logical errors during their reasoning processes. Existing solutions, such as deploying task-specific verifiers or voting over multiple reasoning paths, either require extensive human annotations or fail in scenarios with inconsistent responses. To address these challenges, we introduce RankPrompt, a new prompting method that enables LLMs to self-rank their responses without additional resources. RankPrompt breaks down the ranking problem into a series of comparisons among diverse responses, leveraging the inherent capabilities of LLMs to generate chains of comparison as contextual exemplars. Our experiments across 11 arithmetic and commonsense reasoning tasks show that RankPrompt significantly enhances the reasoning performance of ChatGPT and GPT-4, with improvements of up to 13%. Moreover, RankPrompt excels in LLM-based automatic evaluations for open-ended tasks, aligning with human judgments 74% of the time in the AlpacaEval dataset. It also exhibits robustness to variations in response order and consistency. Collectively, our results validate RankPrompt as an effective method for eliciting high-quality feedback from language models.

Prior research has found that differences in the early period of neural network training significantly impact the performance of in-distribution (ID) tasks. However, neural networks are often sensitive to out-of-distribution (OOD) data, making them less reliable in downstream applications. Yet, the impact of the early training period on OOD generalization remains understudied due to its complexity and lack of effective analytical methodologies. In this work, we investigate the relationship between learning dynamics and OOD generalization during the early period of neural network training. We utilize the trace of Fisher Information and sharpness, with a focus on gradual unfreezing (i.e. progressively unfreezing parameters during training) as the methodology for investigation. Through a series of empirical experiments, we show that 1) selecting the number of trainable parameters at different times during training, i.e. realized by gradual unfreezing -- has a minuscule impact on ID results, but greatly affects the generalization to OOD data; 2) the absolute values of sharpness and trace of Fisher Information at the initial period of training are not indicative for OOD generalization, but the relative values could be; 3) the trace of Fisher Information and sharpness may be used as indicators for the removal of interventions during early period of training for better OOD generalization.

Empathetic response generation endeavors to empower dialogue systems to perceive speakers' emotions and generate empathetic responses accordingly. Psychological research demonstrates that emotion, as an essential factor in empathy, encompasses trait emotions, which are static and context-independent, and state emotions, which are dynamic and context-dependent. However, previous studies treat them in isolation, leading to insufficient emotional perception of the context, and subsequently, less effective empathetic expression. To address this problem, we propose Combining Trait and State emotions for Empathetic Response Model (CTSM). Specifically, to sufficiently perceive emotions in dialogue, we first construct and encode trait and state emotion embeddings, and then we further enhance emotional perception capability through an emotion guidance module that guides emotion representation. In addition, we propose a cross-contrastive learning decoder to enhance the model's empathetic expression capability by aligning trait and state emotions between generated responses and contexts. Both automatic and manual evaluation results demonstrate that CTSM outperforms state-of-the-art baselines and can generate more empathetic responses.