In most classification models, it has been assumed to have a single ground truth label for each data point. However, subjective tasks like toxicity classification can lead to genuine disagreement among annotators. In these cases aggregating labels will result in biased labeling and, consequently, biased models that can overlook minority opinions. Previous studies have shed light on the pitfalls of label aggregation and have introduced a handful of practical approaches to tackle this issue. Recently proposed multi-annotator models, which predict labels individually per annotator, are vulnerable to under-determination for annotators with small samples. This problem is especially the case in crowd-sourced datasets. In this work, we propose Annotator Aware Representations for Texts (AART) for subjective classification tasks. We will show the improvement of our method on metrics that assess the performance on capturing annotators' perspectives. Additionally, our approach involves learning representations for annotators, allowing for an exploration of the captured annotation behaviors.
Large language models (LLMs) have garnered significant attention for their remarkable performance in a continuously expanding set of natural language processing tasks. However, these models have been shown to harbor inherent societal biases, or stereotypes, which can adversely affect their performance in their many downstream applications. In this paper, we introduce a novel, purely prompt-based approach to uncover hidden stereotypes within any arbitrary LLM. Our approach dynamically generates a knowledge representation of internal stereotypes, enabling the identification of biases encoded within the LLM's internal knowledge. By illuminating the biases present in LLMs and offering a systematic methodology for their analysis, our work contributes to advancing transparency and promoting fairness in natural language processing systems.
Large Language Models (LLMs) have seen widespread deployment in various real-world applications. Understanding these biases is crucial to comprehend the potential downstream consequences when using LLMs to make decisions, particularly for historically disadvantaged groups. In this work, we propose a simple method for analyzing and comparing demographic bias in LLMs, through the lens of job recommendations. We demonstrate the effectiveness of our method by measuring intersectional biases within ChatGPT and LLaMA, two cutting-edge LLMs. Our experiments primarily focus on uncovering gender identity and nationality bias; however, our method can be extended to examine biases associated with any intersection of demographic identities. We identify distinct biases in both models toward various demographic identities, such as both models consistently suggesting low-paying jobs for Mexican workers or preferring to recommend secretarial roles to women. Our study highlights the importance of measuring the bias of LLMs in downstream applications to understand the potential for harm and inequitable outcomes.
Causal inference of exact individual treatment outcomes in the presence of hidden confounders is rarely possible. Instead, recent work has adapted conformal prediction to produce outcome intervals. Unfortunately this family of methods tends to be overly conservative, sometimes giving uninformative intervals. We introduce an alternative approach termed Caus-Modens, for characterizing causal outcome intervals by modulated ensembles. Motivated from Bayesian statistics and ensembled uncertainty quantification, Caus-Modens gives tighter outcome intervals in practice, measured by the necessary interval size to achieve sufficient coverage on three separate benchmarks. The last benchmark is a novel usage of GPT-4 for observational experiments with unknown but probeable ground truth.
Pragmatic reference enables efficient interpersonal communication. Prior work uses simple reference games to test models of pragmatic reasoning, often with unidentified speakers and listeners. In practice, however, speakers' sociocultural background shapes their pragmatic assumptions. For example, readers of this paper assume NLP refers to "Natural Language Processing," and not "Neuro-linguistic Programming." This work introduces the Cultural Codes dataset, which operationalizes sociocultural pragmatic inference in a simple word reference game. Cultural Codes is based on the multi-turn collaborative two-player game, Codenames Duet. Our dataset consists of 794 games with 7,703 turns, distributed across 153 unique players. Alongside gameplay, we collect information about players' personalities, values, and demographics. Utilizing theories of communication and pragmatics, we predict each player's actions via joint modeling of their sociocultural priors and the game context. Our experiments show that accounting for background characteristics significantly improves model performance for tasks related to both clue giving and guessing, indicating that sociocultural priors play a vital role in gameplay decisions.
Automatic assessment of the quality of arguments has been recognized as a challenging task with significant implications for misinformation and targeted speech. While real world arguments are tightly anchored in context, existing efforts to judge argument quality analyze arguments in isolation, ultimately failing to accurately assess arguments. We propose SPARK: a novel method for scoring argument quality based on contextualization via relevant knowledge. We devise four augmentations that leverage large language models to provide feedback, infer hidden assumptions, supply a similar-quality argument, or a counterargument. We use a dual-encoder Transformer architecture to enable the original argument and its augmentation to be considered jointly. Our experiments in both in-domain and zero-shot setups show that SPARK consistently outperforms baselines across multiple metrics. We make our code available to encourage further work on argument assessment.
Temporal knowledge graph (TKG) forecasting benchmarks challenge models to predict future facts using knowledge of past facts. In this paper, we apply large language models (LLMs) to these benchmarks using in-context learning (ICL). We investigate whether and to what extent LLMs can be used for TKG forecasting, especially without any fine-tuning or explicit modules for capturing structural and temporal information. For our experiments, we present a framework that converts relevant historical facts into prompts and generates ranked predictions using token probabilities. Surprisingly, we observe that LLMs, out-of-the-box, perform on par with state-of-the-art TKG models carefully designed and trained for TKG forecasting. Our extensive evaluation presents performances across several models and datasets with different characteristics, compares alternative heuristics for preparing contextual information, and contrasts to prominent TKG methods and simple frequency and recency baselines. We also discover that using numerical indices instead of entity/relation names, i.e., hiding semantic information, does not significantly affect the performance ($\pm$0.4\% Hit@1). This shows that prior semantic knowledge is unnecessary; instead, LLMs can leverage the existing patterns in the context to achieve such performance. Our analysis also reveals that ICL enables LLMs to learn irregular patterns from the historical context, going beyond simple predictions based on common or recent information.
Morality plays an important role in culture, identity, and emotion. Recent advances in natural language processing have shown that it is possible to classify moral values expressed in text at scale. Morality classification relies on human annotators to label the moral expressions in text, which provides training data to achieve state-of-the-art performance. However, these annotations are inherently subjective and some of the instances are hard to classify, resulting in noisy annotations due to error or lack of agreement. The presence of noise in training data harms the classifier's ability to accurately recognize moral foundations from text. We propose two metrics to audit the noise of annotations. The first metric is entropy of instance labels, which is a proxy measure of annotator disagreement about how the instance should be labeled. The second metric is the silhouette coefficient of a label assigned by an annotator to an instance. This metric leverages the idea that instances with the same label should have similar latent representations, and deviations from collective judgments are indicative of errors. Our experiments on three widely used moral foundations datasets show that removing noisy annotations based on the proposed metrics improves classification performance.
Warning: this paper contains content that maybe offensive or upsetting. Recent research in Natural Language Processing (NLP) has advanced the development of various toxicity detection models with the intention of identifying and mitigating toxic language from existing systems. Despite the abundance of research in this area, less attention has been given to adversarial attacks that force the system to generate toxic language and the defense against them. Existing work to generate such attacks is either based on human-generated attacks which is costly and not scalable or, in case of automatic attacks, the attack vector does not conform to human-like language, which can be detected using a language model loss. In this work, we propose attacks against conversational agents that are imperceptible, i.e., they fit the conversation in terms of coherency, relevancy, and fluency, while they are effective and scalable, i.e., they can automatically trigger the system into generating toxic language. We then propose a defense mechanism against such attacks which not only mitigates the attack but also attempts to maintain the conversational flow. Through automatic and human evaluations, we show that our defense is effective at avoiding toxic language generation even against imperceptible toxicity triggers while the generated language fits the conversation in terms of coherency and relevancy. Lastly, we establish the generalizability of such a defense mechanism on language generation models beyond conversational agents.
*Content warning: This work displays examples of explicit and strongly offensive language. The COVID-19 pandemic has fueled a surge in anti-Asian xenophobia and prejudice. Many have taken to social media to express these negative sentiments, necessitating the development of reliable systems to detect hate speech against this often under-represented demographic. In this paper, we create and annotate a corpus of Twitter tweets using 2 experimental approaches to explore anti-Asian abusive and hate speech at finer granularity. Using the dataset with less biased annotation, we deploy multiple models and also examine the applicability of other relevant corpora to accomplish these multi-task classifications. In addition to demonstrating promising results, our experiments offer insights into the nuances of cultural and logistical factors in annotating hate speech for different demographics. Our analyses together aim to contribute to the understanding of the area of hate speech detection, particularly towards low-resource groups.