United States

This work aims to assess the zero-shot social reasoning capabilities of LLMs by proposing various strategies based on the granularity of information used to track the fine-grained evolution in the relationship between characters in a book. Without gold annotations, we thoroughly analyze the agreements between predictions from multiple LLMs and manually examine their consensus at a local and global level via the task of trope prediction. Our findings reveal low-to-moderate agreement among LLMs and humans, reflecting the complexity of the task. Analysis shows that LLMs are sensitive to subtle contextual changes and often rely on surface-level cues. Humans, too, may interpret relationships differently, leading to disagreements in annotations.

NAACL 2025

Tracking Evolving Relationship Between Characters in Books in the Era of Large Language Models

long context modeling

evolving relationships

narrative understanding

workshop 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.

Please register!

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

While Large Language Models (LLMs) have been shown to excel at various natural language understanding and reasoning problems, human-level deeper understanding of long-form narratives and stories remains a challenge. In this work, we consider the problem of detecting plot holes in stories: story inconsistencies that goes against the flow of logic established by its plot. We propose an algorithm--PLOTHOLEMAKER-- that creates a high-quality and contamination-free benchmark for this problem by synthesizing plot holes in existing stories. We find that state-of-the-art reasoning models struggle in accurately solving our synthesized task, PLOTHOLEBENCH, that includes careful human-quality filtering. We also observe that scaling the number of reasoning tokens results in little to no improvement on the performance of these models on our benchmark and their performance becomes substantially worse when considering longer stories.

Evaluating LLMs' Ability to Reason about Impossible Worlds

Speaker identification in narrative analysis is a challenging task due to complex dialogues, diverse utterance patterns, and ambiguous character references. Cosly and time-intensive manual annotation limits the scalability of high-quality dataset creation. This study demonstrates a cost-efficient approach of constructing speaker identification datasets by combining small-scale manual annotation with LLM-based labeling. A subset of data is manually annotated and is used to guide LLM predictions with a few-shot approach followed by refinement through minimal human corrections. Our results show that LLMs achieve approximately 90% accuracy on challenging narratives, such as the "Three Kingdoms'' dataset, underscoring the importance of targeted human corrections. This approach proves effective for constructing scalable and cost-efficient datasets for Japanese and complex narratives.

Speaker Identification and Dataset Construction Using LLMs: A Case Study on Japanese Narratives

Books, while often rich in cultural insights, can also mirror societal biases of their eras—biases that Large Language Models (LLMs) may learn and perpetuate during training. We introduce a novel method to trace and quantify these biases using fine-tuned LLMs. We develop BookPAGE, a corpus comprising 593 fictional books across seven decades (1950-2019), to track bias evolution. By fine-tuning LLMs on books from each decade and using targeted prompts, we examine shifts in biases related to gender, sexual orientation, race, and religion. Our findings indicate that LLMs trained on decade-specific books manifest biases reflective of their times, with both gradual trends and notable shifts. For example, model responses showed a progressive increase in the portrayal of women in leadership roles (from 8% to 22%) from the 1950s to 2010s, with a significant uptick in the 1990s (from 4% to 12%), possibly aligning with third-wave feminism. Same-sex relationship references increased markedly from the 1980s to 2000s (from 0% to 10%), mirroring growing LGBTQ+ visibility. Concerningly, negative portrayals of Islam rose sharply in the 2000s (26% to 38%), likely reflecting post-9/11 sentiments. Importantly, we demonstrate that these biases stem mainly from the books' content and not the models' architecture or initial training. Our study offers a new perspective on societal bias trends by bridging AI, literary studies, and social science research.

Fine-Tuned LLMs are "Time Capsules'' for Tracking Societal Bias Through Books

We propose a novel framework to generate causal graphs from narrative texts, bridging the gap between high-level causality and finer-grained event-specific relationships. Our approach first extracts concise, agent-centered "vertices" using an LLM-based summarization strategy. We then introduce an *Expert Index*—seven linguistically grounded features—and incorporate them into a **STAC** (Situation, Task, Action, Consequence) classification model. This hybrid system (RoBERTa embeddings + Expert Index) achieves superior precision in identifying causal links compared to LLM-only baselines. Finally, we apply a structured, five-iteration prompting process to refine and construct a connected causal graph. Experiments on 100 chapters and short stories show that our method consistently outperforms GPT-4o and Claude 3.5 across key dimensions of causal graph quality, while maintaining comparable readability. The resulting open-source tool offers an interpretable and efficient solution for capturing nuanced causal chains within narrative texts.

Beyond LLMs A Linguistic Approach to Causal Graph Generation from Narrative Texts

Interactive storytelling benefits from planning and exploring multiple ``what if'' scenarios. Modern LLMs are useful tools for ideation and exploration, but current chat-based user interfaces restrict users to a single linear flow. To address this limitation, we propose Narrative Studio -- a novel in-browser narrative exploration environment featuring a tree-like interface that allows branching exploration from user-defined points in a story. Each branch is extended via iterative LLM inference guided by system and user-defined prompts. Additionally, we employ Monte Carlo Tree Search (MCTS) to automatically expand promising narrative paths based on user-specified criteria, enabling more diverse and robust story development. We also allow users to enhance narrative coherence by grounding the generated text in a graph that represents the actors and environment of the story.

Narrative Studio: Visual narrative exploration using LLMs and Monte Carlo Tree Search

Recent advancements in language representation learning primarily emphasize language modeling for deriving meaningful representations, often neglecting style-specific considerations. This study addresses this gap by creating generic, sentence-level style embeddings crucial for style-centric tasks. Our approach is grounded on the premise that low-level text style changes can compose any high-level style. We hypothesize that applying this concept to representation learning enables the development of versatile text style embeddings. By fine-tuning a general-purpose text encoder using contrastive learning and standard cross-entropy loss, we aim to capture these low-level style shifts, anticipating that they offer insights applicable to high-level text styles. The outcomes prompt us to reconsider the underlying assumptions as the results do not always show that the learned style representations capture high-level text styles.

Challenging Assumptions in Learning Generic Text Style Embeddings

Sequence-to-sequence models are widely used to train Abstract Meaning Representation (Banarescu et al.,2013, AMR) parsers. To train such models, AMR graphs have to be linearized into a one-line text format. While Penman encoding is widely used for this purpose, we argue that it has limitations: 1) for deep graphs, some closely related nodes are located far apart in the linearized text 2) Penman's tree-based encoding necessitates inverse roles to handle node re-entrancy, doubling the number of relation types to predict. To address these issues, we propose a triple-based linearization method and compare its efficiency by training an AMR parser with both approaches. Although triple is well suited to represent a graph, our results show that it does not yet improve performance on deeper or longer graphs. It suggests room for improvement in its design to better compete with Penman's concise representation and explicit encoding of a nested graph structure.

Reassessing Graph Linearization for Sequence-to-sequence AMR Parsing: On the Advantages and Limitations of Triple-Based

In-context learning (ICL) has transformed the use of large language models (LLMs) for NLP tasks, enabling few-shot learning by conditioning on labeled examples without finetuning. Despite its effectiveness, ICL is prone to errors, especially for challenging examples. With the goal of improving the performance of ICL, we propose *corrective in-context learning* (CICL), an approach that incorporates a model's incorrect predictions alongside ground truth corrections into the prompt, aiming to enhance classification accuracy through self-correction. However, contrary to our hypothesis, extensive experiments on text classification tasks demonstrate that CICL consistently underperforms standard ICL, with performance degrading as the proportion of corrections in the prompt increases. Our findings indicate that CICL introduces confusion by disrupting the model's task understanding, rather than refining its predictions. Additionally, we observe that presenting harder examples in standard ICL does not improve performance, suggesting that example difficulty alone may not be a reliable criterion for effective selection. By presenting these negative results, we provide important insights into the limitations of self-corrective mechanisms in LLMs and offer directions for future research.

Corrective In-Context Learning: Evaluating Self-Correction in Large Language Models

Multilingual large language models (LLMs) aim towards robust natural language understanding across diverse languages, yet their performance significantly degrades on low-resource languages. This work explores whether existing techniques to identify language-specific neurons can be leveraged to enhance cross-lingual task performance of low-resource languages. We conduct detailed experiments covering existing language-specific neuron identification techniques (such as Language
Activation Probability Entropy and activation probability-based thresholding) and
neuron-specific LoRA fine-tuning with models like Llama 3.1 and Mistral Nemo. We find that such neuron-specific interventions are insufficient to yield cross-lingual improvements on downstream tasks (XNLI, XQuAD) in low-resource languages. This study highlights the challenges in achieving cross-lingual generalization and provides critical insights for multilingual LLMs.

Language-Specific Neurons Do Not Facilitate Cross-Lingual Transfer

Intermediate Layer Distillation (ILD) is a variant of Knowledge Distillation (KD), a method for compressing neural networks.
ILD requires mapping to align the intermediate layer sizes of the teacher and student models to compute the loss function in training, while this mapping is not used during inference.
This inconsistency may reduce the effectiveness of learning in intermediate layers.
In this study, we propose LoRAILD, which uses LoRA adapters to eliminate the inconsistency.
However, our experimental results show that LoRAILD does not outperform existing methods.
Furthermore, contrary to previous studies, we observe that conventional ILD does not outperform vanilla KD.
Our analysis of the distilled models' intermediate layers suggests that ILD does not improve language models' performance.

Downloads

Next from NAACL 2025

Evaluating LLMs' Ability to Reason about Impossible Worlds

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

.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

Evaluating LLMs' Ability to Reason about Impossible Worlds

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Downloads