United States

Recent efforts to integrate low-rank adaptation (LoRA) with the Mixture-of-Experts (MoE) have managed to achieve performance comparable to full-parameter fine-tuning by tuning much fewer parameters. Despite promising results, research on improving the efficiency and expert analysis of LoRA with MoE is still in its early stages. Recent studies have shown that experts in the MoE architecture have different strengths and also exhibit some redundancy. Does this statement also apply to parameter-efficient MoE? In this paper, we introduce a novel parameter-efficient MoE method, ***M**oE-L**o**RA with **L**ayer-wise Expert **A**llocation (MoLA)* for Transformer-based models, where each model layer uses a varying number of LoRA experts. We investigate several architectures with varying layer-wise expert configurations. Experiments on six well-known NLP and commonsense QA benchmarks demonstrate that MoLA achieves equal or superior performance compared to all baselines on top of both LLAMA-2, Mistral, and Gemma. We find that allocating more LoRA experts to middle layers further enhances the effectiveness of models with a certain number of experts in total. The redundancy of the experts is more obvious in the lower layers. With much fewer parameters, this allocation strategy outperforms the setting with the same number of experts in every layer. This work can be widely used as a plug-and-play parameter-efficient tuning approach for various applications. The code has been made available at &lt;https://github.com/GCYZSL/MoLA&gt;.

NAACL 2025

MoLA: MoE LoRA with Layer-wise Expert Allocation

layerwise-allocation

parameter-efficient-training

mixture-of-experts

## Welcome to EMNLP 2024! 
We are excited to welcome you to one of the most prominent conferences in the field of Natural Language Processing. This year, EMNLP 2024 is being held in a hybrid format,
offering both virtual and in-person participation in beautiful Miami. Due to a record-breaking number of submissions, we've expanded the total number of accepted papers to accommodate more cutting-edge research from around the globe.
### [Conference Handbook](https://drive.google.com/file/d/1WPROgxjLAC96AJL7Ugy0tEnYm7dkrbHt/view?usp=sharing)

You are required to register for this event. **Please register [here](https://2024.emnlp.org/registration/).** The EMNLP 2024 event page on Underline will be open to public one week prior to the event.

Please register!

EMNLP 2024

EMNLP 2024 will take place in Miami, Florida from Nov 12th to Nov 16th, 2024, at the Hyatt Regency Miami Hote and on Underline for remote participants.

Recent efforts to integrate low-rank adaptation (LoRA) with the Mixture-of-Experts (MoE) have managed to achieve performance comparable to full-parameter fine-tuning by tuning much fewer parameters. Despite promising results, research on improving the efficiency and expert analysis of LoRA with MoE is still in its early stages. Recent studies have shown that experts in the MoE architecture have different strengths and also exhibit some redundancy. Does this statement also apply to parameter-efficient MoE? In this paper, we introduce a novel parameter-efficient MoE method, ***M**oE-L**o**RA with **L**ayer-wise Expert **A**llocation (MoLA)* for Transformer-based models, where each model layer uses a varying number of LoRA experts. We investigate several architectures with varying layer-wise expert configurations. Experiments on six well-known NLP and commonsense QA benchmarks demonstrate that MoLA achieves equal or superior performance compared to all baselines on top of both LLAMA-2, Mistral, and Gemma. We find that allocating more LoRA experts to middle layers further enhances the effectiveness of models with a certain number of experts in total. The redundancy of the experts is more obvious in the lower layers. With much fewer parameters, this allocation strategy outperforms the setting with the same number of experts in every layer. This work can be widely used as a plug-and-play parameter-efficient tuning approach for various applications. The code has been made available at <https://github.com/GCYZSL/MoLA>.

poster

### Welcome to 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics

Welcome to the 2025 meeting of the Nations of the Americas Chapter of the Association for Computational Linguistics! I am proud to help organize the first NAACL conference to carry the new name of our organization, one that emphasizes inclusion for all of the Americas. I am also pleased to welcome you to Albuquerque, New Mexico, a state whose unique blend of cultural influences will make for an excellent backdrop for NAACL 2025, especially with this year’s special theme on NLP in a Multicultural World. 
**[Continue reading...](https://drive.google.com/file/d/1jX-qGhqVSZZCIrAnJaz798pflu5Irrdn/view?usp=sharing)**

*- Colin Cherry, Google, NAACL 2025 General Chair* 

[![](https://assets.underline.io/markdown_image/1/image/b087f8a4dc5816d6e1a6514e59c59ac3.png)](https://drive.google.com/file/d/1T96GzPqObXrMln2BMByCSXTSizjTg69P/view?usp=sharing)

You need to log in with the email address you registered with. Access credentials have been sent to your email. 

Please be sure to check your spam and other email folders if you do not see an email confirmation right away.

Please log in to explore this event.

To access NAACL 2025 event page you are required to register. Please follow [**this link**](https://2025.naacl.org/registration/registration/) to register. Access will depend on your registration type.

Welcome to 2025 Annual Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics

Zero-shot named entity recognition (NER) is the task of detecting named entities of specific types (such as Person or Medicine) without any training examples. Current research increasingly relies on large synthetic datasets, automatically generated to cover tens of thousands of distinct entity types, to train zero-shot NER models. However, in this paper, we find that these synthetic datasets often contain entity types that are semantically highly similar to (or even the same as) those in standard evaluation benchmarks. Because of this overlap, we argue that reported F1 scores for zero-shot NER overestimate the true capabilities of these approaches. Further, we argue that current evaluation setups provide an incomplete picture of zero-shot abilities since they do not quantify the label shift (i.e., the similarity of labels) between training and evaluation datasets. To address these issues, we propose Familarity, a novel metric that captures both the semantic similarity between entity types in training and evaluation, as well as their frequency in the training data, to provide an estimate of label shift. It allows researchers to contextualize reported zero-shot NER scores when using custom synthetic training datasets. Further, it enables researchers to generate evaluation setups of various transfer difficulties for fine-grained analysis of zero-shot NER.

Familarity: Better Evaluation of Zero-Shot Named Entity Recognition by Quantifying Label Shifts in Synthetic Training Data

We propose WHoW, an evaluation framework for analyzing the facilitation strategies of moderators across different domains/scenarios by examining their motives (Why), dialogue acts (How) and target speaker (Who). Using this framework, we annotated 5,657 moderation sentences with human judges and 15,494 sentences with GPT-4o from two domains: TV debates and radio panel discussions. Comparative analysis demonstrates the framework's cross-domain generalisability and reveals distinct moderation strategies: debate moderators emphasise coordination and facilitate interaction through questions and instructions, while panel discussion moderators prioritize information provision and actively participate in discussions. Our analytical framework works for different moderation scenarios, enhances our understanding of moderation behaviour through automatic large-scale analysis, and facilitates the development of moderator agents.

WHoW: A Cross-domain Approach for Analysing Conversation Moderation

Pre-trained Large Language Models (LLMs) have demonstrated remarkable capabilities but also pose risks by learning and generating copyrighted material, leading to significant legal and ethical concerns. In real-world scenarios, model owners need to continuously address copyright infringement as new requests for content removal emerge at different time points. This leads to the need for sequential unlearning, where copyrighted content is removed sequentially as new requests arise. Despite its practical relevance, sequential unlearning in the context of copyright infringement has not been rigorously explored in existing literature. To address this gap, we propose Stable Sequential Unlearning (SSU), a novel framework designed to unlearn copyrighted content from LLMs over multiple time steps. Our approach works by identifying and removing specific weight updates in the model's parameters that correspond to copyrighted content. We improve unlearning efficacy by introducing random labeling loss and ensuring the model retains its general-purpose knowledge by adjusting targeted parameters. Experimental results show that SSU achieves an effective trade-off between unlearning efficacy and general-purpose language abilities, outperforming existing baselines.

Avoiding Copyright Infringement via Large Language Model Unlearning

To reduce the latency associated with autoretrogressive LLM inference, speculative decoding has emerged as a novel decoding paradigm, where future tokens are drafted and verified in parallel. However, the practical deployment of speculative decoding is hindered by its requirements for additional computational resources and fine-tuning, which limits its out-of-the-box usability. To address these challenges, we present PLD+, a suite of novel algorithms developed to accelerate the inference process of LLMs, particularly for input-guided tasks. These tasks, which include code editing, text editing, summarization, etc., often feature outputs with substantial overlap with their inputs—an attribute PLD+ is designed to exploit. PLD+ also leverages the artifacts (attention and hidden states) generated during inference to accelerate inference speed. We test our approach on five input-guided tasks and through extensive experiments we find that PLD+ outperforms all tuning-free approaches. In the greedy setting, it even outperforms the state-of-the-art tuning-dependent approach EAGLE on four of the tasks. (by a margin of upto 2.31 in terms of avg. speedup). Our approach is tuning free, does not require any additional compute and can easily be used for accelerating inference of any LLM.

PLD+: Accelerating LLM Inference by Leveraging Language Model Artifacts

The ability to achieve long-term goals is a key challenge in the current development of large language models (LLMs). To address this, pre-trained LLMs can be fine-tuned with reinforcement learning (RL) to explore solutions that optimize a given goal. However, exploration with LLMs is difficult, as a balance has to be struck between discovering new solutions and staying close enough to the pre-trained model, so as not to degrade basic capabilities. This is typically controlled with a Kullback-Leibler (KL) penalty. In this paper, we investigate the exploration dynamics of a small language model on a simple arithmetic task. We show how varying degrees of pre-training influence exploration and demonstrate the importance of ``critical tokens'' which have a dramatic impact on the final outcome. Consequently, we introduce a simple modification to the KL penalty that favors exploration on critical tokens, increasing the efficiency of the RL fine-tuning stage.

Ignore the KL Penalty! Boosting Exploration on Critical Tokens to Enhance RL Fine-Tuning

Language Model (LM) pruning compresses the model by removing weights, nodes, or other parts of its architecture. Typically, pruning focuses on the resulting efficiency gains at the cost of effectiveness. However, when looking at how individual data points are affected by pruning, it turns out that a particular subset of data points always bears most of the brunt (in terms of reduced accuracy) when pruning, but this effect goes unnoticed when reporting the mean accuracy of all data points. These data points are called PIEs and have been studied in image processing, but not in NLP. In a study of various NLP datasets, pruning methods, and levels of compression, we find that PIEs impact inference quality considerably, regardless of class frequency, and that BERT is more prone to this than BiLSTM. We also find that PIEs contain a high amount of data points that have the largest influence on how well the model generalises to unseen data. This means that when pruning, with seemingly moderate loss to accuracy across all data points, we in fact hurt tremendously those data points that matter the most. We trace what makes PIEs both hard and impactful to inference to their overall longer and more semantically complex text. These findings are novel and contribute to understanding how LMs are affected by pruning. The code is available at: https://github.com/pietrotrope/AsEasyAsPIE

As easy as PIE: understanding when pruning causes language models to disagree

Personalizing large language models (LLMs) is essential for delivering tailored interactions that improve user experience. Many existing personalization methods require fine-tuning LLMs for each user, rendering them prohibitively expensive for widespread adoption. Although retrieval-based approaches offer a more compute-efficient alternative, they still depend on large, high-quality datasets that are not consistently available for all users. To address this challenge, we propose Chameleon, a scalable and efficient personalization approach that uses (1) self-generated personal preference data and (2) representation editing to enable quick and cost-effective personalization. Our experiments on various tasks, including those from the LaMP personalization benchmark, show that Chameleon efficiently adapts models to personal preferences, improving instruction-tuned models and outperforms two personalization baselines by an average of 40% across two model architectures.

Personalize Your LLM: Fake it then Align it

Large Language Models (LLMs) have been shown to be biased in prior work, as they generate text that is in line with stereotypical views of the world or that is not representative of the viewpoints and values of historically marginalized demographic groups. In this work, we propose using data from parallel men's and women's events at the Olympic Games to investigate different forms of gender bias in language models. We define three metrics to measure bias, and find that models are consistently biased against women when the gender is ambiguous in the prompt. In this case, the model frequently retrieves only the results of the men's event with or without acknowledging them as such, revealing pervasive gender bias in LLMs in the context of athletics.

Sports and Women’s Sports: Gender Bias in Text Generation with Olympic Data

To develop high-performing Visual Language Models (VLMs), it is essential to prepare multimodal resources, such as image-text pairs, interleaved data, and instruction data. While multimodal resources for English are abundant, there is a significant lack of corresponding resources for non-English languages, such as Japanese. To address this problem, we take Japanese as a non-English language and propose Japanese multimodal datasets for rapidly developing a Japanese multimodal model. We collect Japanese image-text pairs and interleaved data from web archives and generate Japanese instruction data using an existing large language model and a VLM. Our experimental results show that a VLM trained on these native datasets outperforms those relying on machine-translated content. The resulting VLM, dataset and code used for training is publicly available (Footnote: https://huggingface.co/llm-jp/llm-jp-3-vila-14b, https://github.com/llm-jp/llm-jp-VILA ).

Constructing Multimodal Datasets from Scratch for Rapid Development of a Japanese Visual Language Model

Event relation detection is a fundamental NLP task, leveraged in many downstream applications, whose modeling requires datasets annotated with event relations of various types. However, systematic and complete annotation of these relations is costly and challenging, due to the quadratic number of event pairs that need to be considered. Consequently, many current event relation datasets lack systematicity and completeness.
In response, we introduce EventFull, the first tool that supports consistent, complete and efficient annotation of temporal, causal and coreference relations via a unified and synergetic process.
A pilot study demonstrates that EventFull accelerates and simplifies the annotation process while yielding high inter-annotator agreement.

Downloads

Next from NAACL 2025

Familarity: Better Evaluation of Zero-Shot Named Entity Recognition by Quantifying Label Shifts in Synthetic Training Data

Similar lecture

AlphaExpert: Assigning LoRA Experts Based on Layer Training Quality