Canada

ChatGPT, the first large language model (LLM) with mass adoption, has demonstrated remarkable performance in numerous natural language tasks. Despite its evident usefulness, evaluating ChatGPT&#39;s performance in diverse problem domains remains challenging due to the closed nature of the model and its continuous updates via Reinforcement Learning from Human Feedback (RLHF). We highlight the issue of data contamination in ChatGPT evaluations, with a case study of the task of stance detection. We discuss the challenge of preventing data contamination and ensuring fair model evaluation in the age of closed and continuously trained models.

ACL 2023

Can we trust the evaluation on ChatGPT?

closed models

chatgpt

data leakage

stance detection

### Welcome to ACL 2023, the 61st Annual Meeting of the Association for Computational Linguistics! 
 The conference will be held in Toronto, Canada, July 9-14, 2023. 
Following the succession of the recent conferences in our field, ACL 2023 will adopt a hybrid format.
While the impact of Covid has considerably diminished in terms of traveling, obtaining visas to Canada
entails a very long process. Moreover, the global economic conditions pose challenges for many individuals to travel to conferences. Recognizing these circumstances, we know many participants may not be
able to attend the conference in person. Therefore, we are committed to providing a great virtual platform
so everyone has the opportunity to interact with other participants and enjoy the conference. Based on the
current registered participants, approxiately 30% have chosen to attend the conference virtually. Whether
you join us in person or virtually, we sincerely hope everyone has a remarkable conference experience. 
This General Chair’s message is where I express my gratitude to the many individuals who have made
enormous contributions to the conference over the past year.

Read [**ACL 2023 General Chair's message**](https://docs.google.com/document/d/1WobYM7norbG4dI48s75HfJoD89qgX5a_F-6U8AteLSA/edit?usp=sharing/) in full.

##### **[Conference Handbook](https://2023.aclweb.org/downloads/acl2023-handbook.pdf)**

The Association for Computational Linguistics (ACL) is the premier international scientific and professional society for people working on computational problems involving human language, a field often referred to as either computational linguistics or natural language processing.

ChatGPT, the first large language model (LLM) with mass adoption, has demonstrated remarkable performance in numerous natural language tasks. Despite its evident usefulness, evaluating ChatGPT's performance in diverse problem domains remains challenging due to the closed nature of the model and its continuous updates via Reinforcement Learning from Human Feedback (RLHF). We highlight the issue of data contamination in ChatGPT evaluations, with a case study of the task of stance detection. We discuss the challenge of preventing data contamination and ensuring fair model evaluation in the age of closed and continuously trained models.

workshop paper

As language models continue to be integrated into applications of personal and societal relevance, ensuring these models’ trustworthiness is crucial, particularly with respect to producing consistent outputs regardless of sensitive attributes. Given that first names may serve as proxies for (intersectional) socio-demographic representations, it is imperative to examine the impact of first names on commonsense reasoning capabilities. In this paper, we study whether a model’s reasoning given a specific input differs based on the first names provided. Our underlying assumption is that the reasoning about Alice should not differ from the reasoning about James. We propose and implement a controlled experimental framework to measure the causal effect of first names on commonsense reasoning, enabling us to distinguish between model predictions due to chance and caused by actual factors of interest. Our results indicate that the frequency of first names has a direct effect on model prediction, with less frequent names yielding divergent predictions compared to more frequent names. To gain insights into the internal mechanisms of models that are contributing to these behaviors, we also conduct an in-depth explainable analysis. Overall, our findings suggest that to ensure model robust- ness, it is essential to augment datasets with more diverse first names during the configuration stage.

Examining the Causal Impact of First Names on Language Models: The Case of Social Commonsense Reasoning

Large language models (LLMs) have become mainstream technology with their versatile use cases and impressive performance. Despite the countless out-of-the-box applications, LLMs are still not reliable. A lot of work is being done to improve the factual accuracy, consistency, and ethical standards of these models through fine-tuning, prompting, and Reinforcement Learning with Human Feedback (RLHF), but no systematic analysis of the responses of these models to different categories of statements, or on their potential vulnerabilities to simple prompting changes is available. In this work, we analyze what confuses GPT-3: how the model responds to certain sensitive topics and what effects the prompt wording has on the model response. We find that GPT-3 correctly disagrees with obvious Conspiracies and Stereotypes but makes mistakes with common Misconceptions and Controversies. The model responses are inconsistent across prompts and settings, highlighting GPT-3's unreliability.

Reliability Check: An Analysis of GPT-3's Response to Sensitive Topics and Prompt Wording

Adversarial attack research in natural language processing (NLP) has made significant progress in designing powerful attack methods and defence approaches. However, few efforts have sought to identify which source samples are the most attackable or robust, i.e. can we determine for an unseen target model, which samples are the most vulnerable to an adversarial attack. This work formally extends the definition of sample attackability/robustness for NLP attacks. Experiments on two popular NLP datasets, four state of the art models and four different NLP adversarial attack methods, demonstrate that sample uncertainty is insufficient for describing characteristics of attackable/robust samples and hence a deep learning based detector can perform much better at identifying the most attackable and robust samples for an unseen target model. Nevertheless, further analysis finds that there is little agreement in which samples are considered the most attackable/robust across different NLP attack methods, explaining a lack of portability of attackability detection methods across attack methods.

Sample Attackability in Natural Language Adversarial Attacks

Harmful content detection models tend to have higher false positive rates for content from marginalized groups. In the context of marginal abuse modeling on Twitter, such disproportionate penalization poses the risk of reduced visibility, where marginalized communities lose the opportunity to voice their opinion on the platform. Current approaches to algorithmic harm mitigation, and bias detection for NLP models are often very ad hoc and subject to human bias. We make two main contributions in this paper. First, we design a novel methodology, which provides a principled approach to detecting and measuring the severity of potential harms associated with a text-based model. Second, we apply our methodology to audit Twitter’s English marginal abuse model , which is used for removing amplification eligibility of marginally abusive content. Without utilizing demographic labels or dialect classifiers, we are still able to detect and measure the severity of issues related to the over-penalization of the speech of marginalized communities, such as the use of reclaimed speech, counterspeech, and identity related terms. In order to mitigate the associated harms, we experiment with adding additional true negative examples and find that doing so provides improvements to our fairness metrics without large degradations in model performance.

A Keyword Based Approach to Understanding the Overpenalization of Marginalized Groups by English Marginal Abuse Models on Twitter

Large-scale Pre-Trained Language Models (PTLMs) capture knowledge from massive human-written data which contains latent societal biases and toxic contents. In this paper, we leverage the primary task of PTLMs, i.e., language modeling, and propose a new metric to quantify manifested implicit representational harms in PTLMs towards 13 marginalized demographics. Using this metric, we conducted an empirical analysis of 24 widely used PTLMs. Our analysis provides insights into the correlation between the proposed metric in this work and other related metrics for representational harm. We observe that our metric correlates with most of the gender-specific metrics in the literature. Through extensive experiments, we explore the connections between PTLMs architectures and representational harms across two dimensions: depth and width of the networks. We found that prioritizing depth over width, mitigates representational harms in some PTLMs. Our code and data can be found at https://github.com/microsoft/SafeNLP.

An Empirical Study of Metrics to Measure Representational Harms in Pre-Trained Language Models

We investigate the phenomenon of an LLM's untruthful response using a large set of 220 handcrafted linguistic features. We focus on GPT-3 models and find that the linguistic profiles of responses are similar across model sizes. That is, how varying-sized LLMs respond to given prompts stays similar on the linguistic properties level. We expand upon this finding by training support vector machines that rely only upon the stylistic components of model responses to classify the truthfulness of statements. Though the dataset size limits our current findings, we show the possibility that truthfulness detection is possible without evaluating the content itself. But at the same time, the limited scope of our experiments must be taken into account in interpreting the results.

Linguistic Properties of Truthful Response

Despite the remarkable performances in various applications, machine learning (ML) models could potentially discriminate. They may result in biasness in decision-making, leading to an impact negatively on individuals and society. Recently, various methods have been developed to mitigate biasness and achieve significant performance. Attention mechanisms are a fundamental component of many state-of-the-art ML models and may potentially impact the fairness of ML models. However, how they explicitly influence fairness has yet to be thoroughly explored. In this paper, we investigate how different attention mechanisms affect the fairness of ML models, focusing on models used in Natural Language Processing (NLP) models. We evaluate the performance of fairness of several models with and without different attention mechanisms on widely used benchmark datasets. Our results indicate that the majority of attention mechanisms that have been assessed can improve the fairness performance of Bidirectional Gated Recurrent Unit (BiGRU) and Bidirectional Long Short Term Memory (BiLSTM) in all three datasets regarding religious and gender-sensitive groups, however, with varying degrees of trade-offs in accuracy measures. Our findings highlight the possibility of fairness being affected by adopting specific attention mechanisms in machine learning models for certain datasets.

Debunking Biases in Attention

By accounting for the part-of-speech of words during text-to-text privatization, we find a significant performance boost for downstream tasks of language models.

Guiding Text-to-Text Privatization by Syntax

This paper presents novel experiments shedding light on the shortcomings of current metrics for assessing biases of gender discrimination made by machine learning algorithms on textual data. We focus on the Bios dataset, and our learning task is to predict the occupation of individuals, based on their biography. Such prediction tasks are common in commercial Natural Language Processing (NLP) applications such as automatic job recommendations. We address an important limitation of theoretical discussions dealing with group-wise fairness metrics: they focus on large datasets, although the norm in many industrial NLP applications is to use small to reasonably large linguistic datasets for which the main practical constraint is to get a good prediction accuracy. We then question how reliable are different popular measures of bias when the size of the training set is simply sufficient to learn reasonably accurate predictions. Our experiments sample the Bios dataset and learn more than 200 models on different sample sizes. This allows us to statistically study our results and to confirm that common gender bias indices provide diverging and sometimes unreliable results when applied to relatively small training and test samples. This highlights the crucial importance of variance calculations for providing sound results in this field.

Are fairness metric scores enough to assess discrimination biases in machine learning?

Wikipedia articles are a common source of training data for Natural Language Processing (NLP) research, especially as a source for corpora in languages other than English. However, research has shown that not all Wikipedia editions are produced organically by native speakers, and there are substantial levels of automation and translation activities in the Wikipedia project that could negatively impact the degree to which they truly represent the language and the culture of native speakers. To encourage transparency in the Wikipedia project, Wikimedia Foundation introduced the depth metric as an indication of the degree of collaboration or how frequently users edit a Wikipedia edition’s articles. While a promising start, this depth metric suffers from a few serious problems, like a lack of adequate handling of inflation of edits metric and a lack of full utilization of users-related metrics. In this paper, we propose the DEPTH+ metric, provide its mathematical definitions, and describe how it reflects a better representation of the depth of human collaborativeness. We also quantify the bot activities in Wikipedia and offer a bot-free depth metric after the removal of the bot-created articles and the bot-made edits on the Wikipedia articles.

Downloads

Next from ACL 2023

Examining the Causal Impact of First Names on Language Models: The Case of Social Commonsense Reasoning

Similar lecture

A Systematic Study and Comprehensive Evaluation of ChatGPT on Benchmark Datasets