United States

We investigate in-context temporal biases in attention heads and transformer outputs. Using cognitive science methodologies, we analyze attention scores and outputs of the GPT-2 models of varying sizes. Across attention heads, we observe effects characteristic of human episodic memory, including temporal contiguity, primacy and recency. Transformer outputs demonstrate a tendency toward in-context serial recall. Importantly, this effect is eliminated after the ablation of the induction heads, which are the driving force behind the contiguity effect. Our findings offer insights into how transformers organize information temporally during in-context learning, shedding light on their similarities and differences with human memory and learning.

NAACL 2025

Emergence of Episodic Memory in Transformers: Characterizing Changes in Temporal Structure of Attention Scores During Training

conditional lag probability

episodic memory

transformers

### Welcome!
The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. Our Virtual Poster Sessions will take place online Thursday, August 22, 2024.

You are required to register for this event. **Please register [here](https://2024.aclweb.org/registration). **

If you have already registered, please check your inbox for an email from Underline granting you access to ACL 2024 content.

Please register!

ACL 2024

The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. More information will be announced soon.

technical paper

### 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

In line with the principle of honesty, there has been a growing effort to train large language models (LLMs) to generate outputs containing epistemic markers. However, evaluation in the presence of epistemic markers has been largely overlooked, raising a critical question: Could the use of epistemic markers in LLM-generated outputs lead to unintended negative consequences? To address this, we present EMBER, a benchmark designed to assess the robustness of LLM-judges to epistemic markers in both single and pairwise evaluation settings. Our findings, based on evaluations using **EMBER**, reveal that all tested LLM-judges, including GPT-4o, show a notable lack of robustness in the presence of epistemic markers. Specifically, we observe a negative bias toward epistemic markers, with a stronger bias against markers expressing uncertainty. This suggests that LLM-judges are influenced by the presence of these markers and do not focus solely on the correctness of the content.

Are LLM-Judges Robust to Expressions of Uncertainty? Investigating the effect of Epistemic Markers on LLM-based Evaluation

Automatic question generation (QG) is essential for AI and NLP, particularly in intelligent tutoring, dialogue systems, and fact verification. Generating multiple-choice questions (MCQG) for professional exams, like the United States Medical Licensing Examination (USMLE), is particularly challenging, requiring domain expertise and complex multi-hop reasoning for high-quality questions. However, current large language models (LLMs) like GPT-4 struggle with professional MCQG due to outdated knowledge, hallucination issues, and prompt sensitivity, resulting in unsatisfactory quality and difficulty. To address these challenges, we propose MCQG-SRefine, an LLM self-refine-based (Critique and Correction) framework for converting medical cases into high-quality USMLE-style questions. By integrating expert-driven prompt engineering with iterative self-critique and self-correction feedback, MCQG-SRefine significantly enhances human expert satisfaction regarding both the quality and difficulty of the questions. Furthermore, we introduce an LLM-as-Judge-based automatic metric to replace the complex and costly expert evaluation process, ensuring reliable and expert-aligned assessments.

MCQG-SRefine: Multiple Choice Question Generation and Evaluation with Iterative Self-Critique, Correction, and Comparison Feedback

Large language models (LLMs) have replaced the metaphorical monkeys in the ”infinite monkeys” thought experiment with machines that mirror human writing. With LLMs being used to generate content at an unprecedented scale, concerns over their misuse and the saturation of the content space with artificially generated material are growing. We foresee a point in the future where a vast majority of all the possible text in a given language would have already been generated,
leading to a ”Plagiarism Singularity”. In this paper, we provide predictions on how far we are from this singularity in the form of an estimate of the volume of content that needs to be generated to reach this singularity. We use an LLM to calculate the probability distribution of sentences in the English language collected from a large dataset. We then estimate the minimum number of sentences to be generated to cover different percentiles of the probability mass of the set of all sentences, assuming they follow the calculated distribution, by treating the problem as an
instance of the coupon collector’s problem. We find that breaching the standard 20% plagiarism limit would only need around 1030 sentences to be generated, which we estimate to happen in approximately 40 years from now.

The Plagiarism Singularity Conjecture

Vision-language (VL) models have demonstrated strong performance across various tasks. However, these models often rely on a specific modality for predictions, leading to ''dominant modality bias.'' This bias significantly hurts performance, especially when one modality is impaired. In this study, we analyze model behavior under dominant modality bias and theoretically show that unaligned gradients or differences in gradient magnitudes prevent balanced convergence of the loss. Based on these findings, we propose a novel framework, **BalGrad** to mitigate dominant modality bias. Our approach includes inter-modality gradient reweighting, adjusting the gradient of KL divergence based on each modality's contribution, and inter-task gradient projection to align task directions in a non-conflicting manner. Experiments on UPMC Food-101, Hateful Memes, and MM-IMDb datasets confirm that **BalGrad** effectively alleviates over-reliance on specific modalities when making predictions.

See-Saw Modality Balance: See Gradient, and Sew Impaired Vision-Language Balance to Mitigate Dominant Modality Bias

Large Language Models (LLMs) have shown promising results in a variety of literary tasks, often using complex memorized details of narration and fictional characters. In this work, we evaluate the ability of Llama-3 at attributing utterances of direct-speech to their speaker in novels. The LLM shows impressive results on a corpus of 28 novels, surpassing published results with ChatGPT and encoder-based baselines by a large margin. We then validate these results by assessing the impact of book memorization and annotation contamination. We found that these types of memorization do not explain the large performance gain, making Llama-3 the new state-of-the-art for quotation attribution in English literature. We release publicly our code and data.

Evaluating LLMs for Quotation Attribution in Literary Texts: A Case Study of LLaMa3

Online reviews provide valuable insights into the perceived quality of facets of a product or service. While aspect-based sentiment analysis has focused on extracting these facets from reviews, there is less work understanding the impact of each aspect on overall perception. This is particularly challenging given correlations among aspects, making it difficult to isolate the effects of each. This paper introduces a methodology based on recent advances in text-based causal analysis, specifically CausalBERT, to disentangle the effect of each factor on overall review ratings. We enhance CausalBERT with three key improvements: temperature scaling for better calibrated treatment assignment estimates; hyperparameter optimization to reduce confound overadjustment; and interpretability methods to characterize discovered confounds. In this work, we treat the textual mentions in reviews as proxies for real-world attributes. We validate our approach on real and semi-synthetic data from over 600K reviews of U.S. K-12 schools. We find that the proposed enhancements result in more reliable estimates, and that perception of school administration and performance on benchmarks are significant drivers of overall school ratings.

Using Text-Based Causal Inference to Disentangle Factors Influencing Online Review Ratings

We analyse the cross-lingual transferability of a dialogue evaluation framework that assesses the relationships between micro-level linguistic features (e.g. backchannels) and macro-level interactivity labels (e.g. topic management), originally designed for English-as-a-second-language dialogues. To this end, we develop CNIMA (**C**hinese **N**on-Native **I**nteractivity **M**easurement and **A**utomation), a Chinese-as-a-second-language labelled dataset with 10K dialogues. We found the evaluation framework to be robust across languages, revealing language-specific and language-universal relationships between micro-level and macro-level features. Next, we propose an automated, interpretable approach with low data requirements that scores the overall quality of a second-language dialogue based on the framework. Our approach is interpretable in that it reveals the key linguistic and interactivity features that contributed to the overall quality score. As our approach does not require labelled data, it can also be adapted to other languages for second-language dialogue evaluation.

An Interpretable and Crosslingual Method for Evaluating Second-Language Dialogues

This paper introduces the novel task of multimodal puzzle solving, framed within the context of visual question-answering. We present a new dataset, AlgoPuzzleVQA designed to challenge and evaluate the capabilities of multimodal language models in solving algorithmic puzzles that necessitate both visual understanding, language understanding, and complex algorithmic reasoning. We create the puzzles to encompass a diverse array of mathematical and algorithmic topics such as boolean logic, combinatorics, graph theory, optimization, search, etc., aiming to evaluate the gap between visual data interpretation and algorithmic problem-solving skills. The dataset is generated automatically from code authored by humans. All our puzzles have exact solutions that can be found from the algorithm without tedious human calculations. It ensures that our dataset can be scaled up arbitrarily in terms of reasoning complexity and dataset size. Our investigation reveals that multimodal language models such as GPT4V and Gemini exhibit limited performance in puzzle-solving tasks. We find that their performance is near random in a multi-choice question-answering setup for a significant number of puzzles. The findings emphasize the challenges of integrating visual, language, and algorithmic knowledge for solving complex reasoning problems.

AlgoPuzzleVQA: Diagnosing Multimodal Reasoning Challenges of Language Models with Algorithmic Multimodal Puzzles

Pre-trained language models form the basis for much of modern NLP, and while the basic pre-training recipe is well known, many details of model development are hidden in secretive research labs. Based on experience training Meta’s Llama models, I will shed light on the evolving science of pre-training. The central challenge of pre-training research is in designing small scale experiments that can confidently be extrapolated to main training runs at orders of magnitude greater scale - which has led to an approach that differs from much of academic research. While careful experiments can disentangle many of the factors that do and don’t matter for large scale training, I will also discuss gaps in the science that mean researcher judgements and intuitions remain critical to decision making.

Downloads

Next from NAACL 2025

Are LLM-Judges Robust to Expressions of Uncertainty? Investigating the effect of Epistemic Markers on LLM-based Evaluation

Similar lecture

Identifying Semantic Induction Heads to Understand In-Context Learning

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from NAACL 2025

Are LLM-Judges Robust to Expressions of Uncertainty? Investigating the effect of Epistemic Markers on LLM-based Evaluation

Similar lecture

Identifying Semantic Induction Heads to Understand In-Context Learning

Downloads