Singapore

Previous studies have established that pre-trained language models inherently manifest various biases captured from the training corpora. Though several debiasing strategies, such as fine-tuning the whole model with counterfactuals, prompt tuning, and representation projection, have been introduced, they often fall short of effectively altering the models&#39; intrinsic parameters without retaining language modeling ability. To address this, we propose EditBias, changing models&#39; biased properties by modifying partial parameters while preserving meaningless associations. Experiments show that EditBias achieves excellent performance on debiasing.

EMNLP 2023

Debiasing Stereotyped Models via Model Editing

bias unlearning

model editing

debiasing

bias

Previous studies have established that pre-trained language models inherently manifest various biases captured from the training corpora. Though several debiasing strategies, such as fine-tuning the whole model with counterfactuals, prompt tuning, and representation projection, have been introduced, they often fall short of effectively altering the models' intrinsic parameters without retaining language modeling ability. To address this, we propose EditBias, changing models' biased properties by modifying partial parameters while preserving meaningless associations. Experiments show that EditBias achieves excellent performance on debiasing.

workshop paper

**Welcome to EMNLP 2023!**

On behalf of the EMNLP 2023 Organizing Committee, I extend a warm and heartfelt welcome to all of you to the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP). It is with immense pleasure and excitement that we gather here in Singapore, a vibrant hub of innovation and technological advancement.

The conference program is packed with insightful presentations, thought-provoking workshops, and engaging networking opportunities. In addition to the technical sessions, we have also planned several social events that will provide you with the opportunity to connect with your colleagues.

The conference is held in person at the Resorts World Convention Centre in Singapore, and available on-line with the help of Underline.

For those in Singapore, I hope that you will find time to explore the exciting Sentosa Island, Gardens by the Bay, Singapore Botanic Gardens and the many other attractions unique to Singapore.

Of course, we are grateful to our sponsors and partners for their generous support of EMNLP 2023, whose contributions make it possible for us to host this world-class event.

Yuji Matsumoto (RIKEN AIP) 
EMNLP 2023 General Chair

To access the **EMNLP 2023** event page on Underline, you need to register for the Conference. 
Please follow **[this link](https://2023.emnlp.org/registration/)** for more details.

EMNLP 2023 took place in Singapore from Dec 6th to Dec 10th, 2023.

Large Language Models (LLMs) have showcased remarkable advancements in diverse language tasks. However, their performance in mathematical reasoning, particularly in solving Mathematical Word Problems (MWPs), remains a challenge. To address this, we propose a novel masking technique tailored to enhance LLMs' MWPs-solving abilities. Unlike random masking, our method employs a balanced and strategic approach, improving the model's discrimination between word tokens and numerical values. This enables LLMs to better understand MWPs intricacies, resulting in significant performance improvements. Our approach bridges the gap between linguistic comprehension and mathematical problem-solving, boosting LLMs' efficacy in solving MWPs and holds potential for broader applications in various knowledge domains.

Premium content

Debiasing Stereotyped Models via Model Editing

Next from EMNLP 2023

Numerical Masking: Enhancing Numerical Proficiency in Language Models for Mathematical Word Problem Solving