Canada

Natural language processing models tend to learn and encode social biases present in the data. One popular approach for addressing such biases is to eliminate encoded information from the model&#39;s representations. However, current methods are restricted to removing only linearly encoded information. In this work, we propose Iterative Gradient-Based Projection (IGBP), a novel method for removing non-linear encoded concepts from neural representations. Our method consists of iteratively training neural classifiers to predict a particular attribute we seek to eliminate, followed by a projection of the representation on a hypersurface, such that the classifiers become oblivious to the target attribute. We evaluate the effectiveness of our method on the task of removing gender and race information as sensitive attributes. Our results demonstrate that IGBP is effective in mitigating bias through intrinsic and extrinsic evaluations, with minimal impact on downstream task accuracy.

ACL 2023

Shielded Representations: Protecting Sensitive Attributes Through Iterative Gradient-Based Projection

concept removal

fairness

bias mitigation

### 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.

Natural language processing models tend to learn and encode social biases present in the data. One popular approach for addressing such biases is to eliminate encoded information from the model's representations. However, current methods are restricted to removing only linearly encoded information. In this work, we propose Iterative Gradient-Based Projection (IGBP), a novel method for removing non-linear encoded concepts from neural representations. Our method consists of iteratively training neural classifiers to predict a particular attribute we seek to eliminate, followed by a projection of the representation on a hypersurface, such that the classifiers become oblivious to the target attribute. We evaluate the effectiveness of our method on the task of removing gender and race information as sensitive attributes. Our results demonstrate that IGBP is effective in mitigating bias through intrinsic and extrinsic evaluations, with minimal impact on downstream task accuracy.

workshop paper

TL;DR: To keep up with the rapid language model development, we propose enlisting the models themselves to help construct bias auditing benchmarks that remain challenging. We apply our methodology to a bias NLI benchmark and create a new, much larger benchmark that uncovers considerably more bias in fine-tuned NLI models. The accuracy of the NLI fine-tuned models, correlated with bias, decreases from 95.3% for the original benchmark to a strikingly low 58.6% for the proposed benchmark. We introduce bias measures that account for different types of model errors and distinguish among pro-/anti-stereotype bias and model error due to model brittleness.


Keeping Up with the Language Models: Robustness-Bias Interplay in NLI Data and Models

Many NLP classification tasks, such as sexism/racism detection or toxicity detection, are based on human values. Yet, human values can vary under diverse cultural conditions. Therefore, we introduce a framework for value-aligned classification that performs prediction based on explicitly written human values in the command. 
Along with the task, we propose a practical approach that distills value-aligned knowledge from large-scale language models (LLMs) to construct value-aligned classifiers in two steps. First, we generate value-aligned training data from LLMs by prompt-based few-shot learning. Next, we fine-tune smaller classification models with the generated data for the task. Empirical results show that our VA-Models surpass multiple baselines by at least 15.56% on the F1-score, including few-shot learning with OPT-175B and existing text augmentation methods. We suggest that using classifiers with explicit human value input improves both inclusivity & explainability in AI.

Enabling Classifiers to Make Judgements Explicitly Aligned with Human Values

Transformer architectures are complex and their use in NLP, while it has engendered many successes, makes their interpretability or explainability challenging. Recent debates have shown that attention maps and attribution methods are unreliable (Pruthi et al., 2019; Brunner et al., 2019). In this paper, we present some of their limitations and introduce COCKATIEL, which successfully addresses some of them. COCKATIEL is a novel, post-hoc, concept-based, model-agnostic XAI technique that generates meaningful explanations from the last layer of a neural net model trained on an NLP classification task by using Non-Negative Matrix Factorization (NMF) to discover the concepts the model leverages to make predictions and by exploiting a Sensitivity Analysis to estimate accurately the importance of each of these concepts for the model. It does so without compromising the accuracy of the underlying model or requiring a new one to be trained. We conduct experiments in single and multi-aspect sentiment analysis tasks and we show COCKATIEL's superior ability to discover concepts that align with humans' on Transformer models without any supervision, we objectively verify the faithfulness of its explanations through fidelity metrics, and we showcase its ability to provide meaningful explanations in two different datasets.

COCKATIEL: COntinuous Concept ranKed ATtribution with Interpretable ELements for explaining neural net classifiers on NLP tasks

Growing literature has shown that NLP systems may encode social biases; however, the political bias of summarization models remains relatively unknown. In this work, we use an entity replacement method to investigate the portrayal of politicians in automatically generated summaries of news articles. We develop a computational framework to assess the sensitivities of several extractive and abstractive summarizers to the political entities Donald Trump and Joe Biden. We find consistent differences in these summaries upon entity replacement, such as reduced emphasis of Trump's presence in the context of the same article and a more individualistic representation of Trump with respect to the collective US government (i.e., administration). Our characterization provides a foundation for future studies of bias in summarization and for normative discussions on the ideal qualities of automatic summaries.


Characterizing Political Bias in Automatic Summaries: A Case Study of Trump and Biden

Textual backdoor attack, as a novel attack model, has been shown to be effective in adding a backdoor to the model during training. Defending against such backdoor attacks has become urgent and important. In this paper, we propose AttDef, an efficient attribution-based pipeline to defend against two insertion-based poisoning attacks, BadNL and InSent Specifically, we regard the tokens with larger attribution scores as potential triggers since larger attribution words contribute more to the false prediction results and therefore are more likely to be poison triggers. Additionally, we further utilize an external pre-trained language model to distinguish whether the input is poisoned or not. We show that our proposed method can generalize sufficiently well in two common attack scenarios (poisoning training data and testing data), which consistently improves previous methods. For instance, AttDef can successfully mitigate both attacks with an aver age accuracy of 79.97% (56.59%↑) and 48.34% (3.99%↑) under pre-training and post-training attack defense respectively, achieving the new state-of-the-art performance on prediction recovery over four benchmark datasets.

Defending against Insertion-based Textual Backdoor Attacks via Attribution

It is common sense that one should prefer to eat a salad with a fork rather than with a chainsaw. However, for eating a bowl of rice, the choice between a fork and a pair of chopsticks is culturally relative. We introduce FORK, a small, manually-curated set of CommonsenseQA-style questions for probing cultural biases and assumptions present in commonsense reasoning systems, with a specific focus on food-related customs. We test several CommonsenseQA systems on FORK, and while we see high performance on questions about the US culture, the poor performance of these systems on questions about non-US cultures highlights systematic cultural assumptions aligned with US over non-US cultures.

FORK: A Bite-Sized Test Set for Probing Culinary Cultural Biases in Commonsense Reasoning Models

Rumour detection on social media is an important topic due to the challenges of misinformation propagation and slow verification of misleading information. Most previous work focus on the response posts on social media, ignoring the useful characteristics of involved users and their relations. In this paper, we propose a novel framework, Post-User Fusion Network (PESTO), which models the patterns of rumours from both post diffusion and user social networks. Specifically, we propose a novel Chronologically-masked Transformer architecture to model both temporal sequence and diffusion structure of rumours, and apply a Relational Graph Convolutional Network to model the social relations of involved users, with a fusion network based on self-attention mechanism to incorporate the two aspects. Additionally, two data augmentation techniques are leveraged to improve the robustness and accuracy of our models. Empirical results on four datasets of English tweets show the superiority of the proposed method.

PESTO: A Post-User Fusion Network for Rumour Detection on Social Media

The task of predicting reader appreciation or literary quality has been the object of several studies. It remains, however, a challenging problem in quantitative literary analyses and computational linguistics alike, as its definition can vary a lot depending on the genre of liter- ary texts considered, the features adopted, and the annotation system employed. This paper attempts to evaluate the impact on reader ap- preciation, defined as online users’ ratings, of sentiment range and sentiment arc patterns ver- sus traditional stylometric features. We run our experiments on a corpus of English-language literary fiction, showing that stylometric fea- tures alone are helpful in modelling literary quality, but can be outperformed by analysing the novels’ sentimental profile.

Sentimental Matters - Predicting Literary Quality by Sentiment Analysis and Stylometric Features

Aspect-based Sentiment Analysis (ABSA) is a fine-grained sentiment analysis task which involves four elements from user-generated texts: aspect term, aspect category, opinion term, and sentiment polarity. Most computational approaches focus on some of the ABSA sub-tasks such as tuple (aspect term, sentiment polarity) or triplet (aspect term, opinion term, sentiment polarity) extraction using either pipeline or joint modeling approaches. Recently, generative approaches have been proposed to extract all four elements as (one or more) quadruplets from text as a single task. In this work, we take a step further and propose a unified framework for solving ABSA, and the associated sub-tasks to improve the performance in few-shot scenarios. To this end, we fine-tune a T5 model with instructional prompts in a multi-task learning fashion covering all the sub-tasks, as well as the entire quadruple prediction task. In experiments with multiple benchmark datasets, we show that the proposed multi-task prompting approach brings performance boost (by absolute 8.29 F1) in the few-shot learning setting.

Instruction Tuning for Few-Shot Aspect-Based Sentiment Analysis

In this work we use consumed text to infer Big-5 personality inventories using data we have collected from the social media platform Reddit. We test our models on two datasets, sampled from participants who consumed either fiction content (N = 913) or news content (N = 213). We show that state-of-the-art models from a similar task using authored text do not translate well to this task, with average correlations of r=.06 between the model's predictions and ground-truth personality inventory dimensions. We propose an alternate method of generating average personality labels for each piece of text consumed, under which our model achieves correlations as high as r=.34 when predicting personality from the text being read.

Next from ACL 2023

Keeping Up with the Language Models: Robustness-Bias Interplay in NLI Data and Models

Similar lecture

Shielded Representations: Protecting Sensitive Attributes Through Iterative Gradient-Based Projection