Austria

Modern language models are typically trained over subword sequences, but ultimately define probabilities over character-strings. Ideally, the choice of the tokeniser---which maps character-strings to subwords---should not affect the probability assigned to the underlying character-string; in practice, it does. We define this mismatch as **tokenisation bias**. In this work, we quantify one particular type of tokenisation bias: the effect of including or not a subword (e.g., $\langle$ hello $\rangle$) in a tokeniser’s vocabulary on the probability a trained model assigns to the corresponding characters (i.e., &quot;hello&#39;&#39;). Estimating this effect is challenging because each model is trained with only one tokeniser. We address this by framing tokenisation bias as a causal effect and estimating it using the regression discontinuity design. Specifically, we exploit the fact that tokenisation algorithms rank subwords and add the first $K$ to a tokeniser&#39;s vocabulary, where $K$ is an arbitrary cutoff point. As such, we can estimate a causal effect by comparing similar subwords around this cutoff. Experimentally, we find that tokenisation consistently affects models&#39; outputs across scales, vocabularies, and tokenisers. Notably, a subword&#39;s presence in a small model&#39;s vocabulary may increase its characters&#39; probability by up to 17 times, highlighting tokenisation as a key design choice in language modelling.

ACL 2025

Causal Estimation of Tokenisation Bias

tokenization

### Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Message from the General Chair: 
*It is my great pleasure and honor to welcome you to the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025), held in beautiful Vienna, Austria, from July 27 to August 1, 2025. ACL2025continues our field’s tradition of excellence in scholarship, innovation, and inclusivity, and I am deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.* 
[Read more](https://drive.google.com/file/d/1GI_hvOpjswAuYdUTromfeDiPpCcqidwg/view?usp=sharing)

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to ACL 2025". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you need to log in with the **email address you registered with**. 

Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Modern language models are typically trained over subword sequences, but ultimately define probabilities over character-strings. Ideally, the choice of the tokeniser---which maps character-strings to subwords---should not affect the probability assigned to the underlying character-string; in practice, it does. We define this mismatch as **tokenisation bias**. In this work, we quantify one particular type of tokenisation bias: the effect of including or not a subword (e.g., $\langle$ hello $\rangle$) in a tokeniser’s vocabulary on the probability a trained model assigns to the corresponding characters (i.e., "hello''). Estimating this effect is challenging because each model is trained with only one tokeniser. We address this by framing tokenisation bias as a causal effect and estimating it using the regression discontinuity design. Specifically, we exploit the fact that tokenisation algorithms rank subwords and add the first $K$ to a tokeniser's vocabulary, where $K$ is an arbitrary cutoff point. As such, we can estimate a causal effect by comparing similar subwords around this cutoff. Experimentally, we find that tokenisation consistently affects models' outputs across scales, vocabularies, and tokenisers. Notably, a subword's presence in a small model's vocabulary may increase its characters' probability by up to 17 times, highlighting tokenisation as a key design choice in language modelling.

poster

Recent research in retrieval-augmented generation (RAG) has concentrated on retrieving useful information from candidate documents. However, numerous methodologies frequently neglect the calibration capabilities of large language models (LLMs), which capitalize on their robust in-context reasoning prowess. This work illustrates that providing LLMs with specific cues substantially improves their calibration efficacy, especially in multi-round calibrations. We present a new SGIC: Self-Guided Iterative Calibration Framework that employs uncertainty scores as a tool. Initially, this framework calculates uncertainty scores to determine both the relevance of each document to the query and the confidence level in the responses produced by the LLMs. Subsequently, it reevaluates these scores iteratively, amalgamating them with prior responses to refine calibration. Furthermore, we introduce an innovative approach for constructing an iterative self-calibration training set, which optimizes LLMs to efficiently harness uncertainty scores for capturing critical information and enhancing response accuracy. Our proposed framework significantly improves performance on both closed-source and open-source LLMs.

SGIC: A Self-Guided Iterative Calibration Framework for RAG

We introduce MedGraphRAG, a novel graph-based Retrieval-Augmented Generation (RAG) framework designed to enhance LLMs in generating evidence-based medical responses, improving safety and reliability with private medical data. We introduce Triple Graph Construction and U-Retrieval to enhance GraphRAG, enabling holistic insights and evidence-based response generation for medical applications. Specifically, we connect user documents to credible medical sources and integrate Top-down Precise Retrieval with Bottom-up Response Refinement for balanced context awareness and precise indexing. Validated on 9 medical Q&A benchmarks, 2 health fact-checking datasets, and a long-form generation test set, MedGraphRAG outperforms state-of-the-art models while ensuring credible sourcing. Our code is publicly available.

Medical Graph RAG: Evidence-based Medical Large Language Model via Graph Retrieval-Augmented Generation

We introduce Agentic Reasoning, a framework that enhances large language model (LLM) reasoning by integrating external tool-using agents. Agentic Reasoning dynamically leverages web search, code execution, and structured memory to address complex problems requiring deep research. A key innovation in our framework is the Mind-Map agent, which constructs a structured knowledge graph to store reasoning context and track logical relationships, ensuring coherence in long reasoning chains with extensive tool usage. Additionally, we conduct a comprehensive exploration of the Web-Search agent, leading to a highly effective search mechanism that surpasses all prior approaches. When deployed on DeepSeek-R1, our method achieves a new state-of-the-art (SOTA) among public models and delivers performance comparable to OpenAI Deep Research, the leading proprietary model in this domain. Extensive ablation studies validate the optimal selection of agentic tools and confirm the effectiveness of our Mind-Map and Web-Search agents in enhancing LLM reasoning. Our code and data are publicly available.

Agentic Reasoning: A Streamlined Framework for Enhancing LLM Reasoning with Agentic Tools

In this work, we present a manually annotated corpus for Adverse Event (AE) extraction from discharge summaries of elderly patients, a population often underrepresented in clinical NLP resources. The dataset includes 14 clinically significant AEs—such as falls, delirium, and intracranial haemorrhage, along with contextual attributes like negation, diagnosis type, and in-hospital occurrence. Uniquely, the annotation schema supports both discontinuous and overlapping entities, addressing challenges rarely tackled in prior work. We evaluate multiple models using FlairNLP across three annotation granularities: fine-grained, coarse-grained, and coarse-grained with negation. While transformer-based models (e.g., BERT-cased) achieve strong performance on document-level coarse-grained extraction (F1 = 0.943), performance drops notably for fine-grained entity-level tasks (e.g., F1 = 0.675), particularly for rare events and complex attributes. These results demonstrate that despite high-level scores, significant challenges remain in detecting underrepresented AEs and capturing nuanced clinical language. Developed within a Trusted Research Environment (TRE), the dataset is available upon request via DataLoch and serves as a robust benchmark for evaluating AE extraction methods and supporting future cross-dataset generalisation.

Adverse Event Extraction from Discharge Summaries: A New Dataset, Annotation Scheme, and Initial Findings

Historical documents represent an invaluable cultural heritage, yet have undergone significant degradation over time through tears, water erosion, and oxidation. Existing Historical Document Restoration (HDR) methods primarily focus on single modality or limited-size restoration, failing to meet practical needs. To fill this gap, we present a full-page HDR dataset (FPHDR) and a novel automated HDR solution (AutoHDR). Specifically, FPHDR comprises 1,633 real and 6,543 synthetic images with character-level and line-level locations, as well as character annotations in different damage grades. AutoHDR mimics historians' restoration workflows through a three-stage approach: OCR-assisted damage localization, vision-language context text prediction, and patch autoregressive appearance restoration. The modular architecture of AutoHDR enables seamless human-machine collaboration, allowing for flexible intervention and optimization at each restoration stage. Experiments demonstrate AutoHDR's remarkable performance in HDR. When processing severely damaged documents, our system improves OCR accuracy from 46.83% to 84.05%, with further enhancement to 94.25% through human-machine collaboration. We believe this work represents a significant advancement in automated historical document restoration and contributes substantially to cultural heritage preservation. The model and dataset are available at https://github.com/SCUT-DLVCLab/AutoHDR.

Reviving Cultural Heritage: A Novel Approach for Comprehensive Historical Document Restoration

Large language models (LLMs) have demonstrated remarkable multilingual capabilities, however, how to evaluate cross-lingual alignment remains underexplored. Existing alignment benchmarks primarily focus on sentence embeddings, but prior research has shown that neural models tend to induce a non-smooth representation space, which impact of semantic alignment evaluation on low-resource languages. Inspired by neuroscientific findings that similar information activates overlapping neuronal regions, we propose a novel *Neuron State-Based Cross-Lingual Alignment* $(\textbf{NeuronXA})$ to assess the cross-lingual a lignment capabilities of LLMs, which offers a more semantically grounded approach to assess cross-lingual alignment. We evaluate NeuronXA on several prominent multilingual LLMs (LLaMA, Qwen, Mistral, GLM, and OLMo) across two transfer tasks and three multilingual benchmarks. The results demonstrate that with only 100 parallel sentence pairs, NeuronXA achieves a Pearson correlation of 0.9556 with downstream tasks performance and 0.8524 with transferability. These findings demonstrate NeuronXA's effectiveness in assessing both cross-lingual alignment and transferability, even with a small dataset. This highlights its potential to advance cross-lingual alignment research and to improve the semantic understanding of multilingual LLMs.

From Neurons to Semantics: Evaluating Cross-Linguistic Alignment Capabilities of Large Language Models via Neurons Alignment

Attributed text generation aims to enhance the reliability of content generated from large language models by providing citations for each claim, which thereby enables users to easily verify the correctness of the responses.
However, the scarcity of high-quality training samples presents a significant challenge in aligning large language models to generate texts with citations, revealing considerable room for improvement in existing attribution systems.
Besides, existing approaches of aligning large language models to follow user instructions can lead to an undue emphasis on irrelevant documents, which in turn reduces the quality of responses.
To address the above problems, we propose \textbf{A}utomatic \textbf{A}lignment Framework for \textbf{A}ttributed Text Generation~(\textbf{$\mathcal{A}^3$}), a novel framework designed to automatically generate high-quality attributed query-response pairs for both supervised fine-tuning and preference optimization stages without human annotation.
With the help of \textbf{$\mathcal{A}^3$}, Mistral-7B can achieve a citation recall of \textbf{84.4} and a precision of \textbf{87.0} precision on ASQA, which notably surpasses GPT-4's citation recall of \textbf{73.0} and precision of \textbf{76.5}.

A3: Automatic Alignment Framework for Attributed Text Generation

Large language models (LLMs) exihibit increasingly sophisticated linguistic capabilities, yet the extent to which these behaviors reflect human-like cognition versus advanced pattern recognition remains an open question.
In this study, we investigate how LLMs process the temporal meaning of linguistic aspect in narratives that were previously used in human studies. Using an Expert-in-the-Loop probing pipeline, we conduct a series of targeted experiments to assess whether LLMs construct semantic representations and pragmatic inferences in a human-like manner.
Our findings show that LLMs over-rely on prototypicality, produce inconsistent aspectual judgments, and struggle with causal reasoning derived from aspect, raising concerns about their ability to fully comprehend narratives.
These results suggest that LLMs process aspect fundamentally differently from humans and lack robust narrative understanding.
Beyond these empirical findings, we develop a standardized experimental framework for the reliable assessment of LLMs' cognitive and linguistic capabilities.

How LLMs Comprehend Temporal Structure in Narratives: A Case Study in Cognitive Evaluation of LLMs

Some statements have one well-defined continuation (e.g., "the Eiffel Tower is in [$Paris$]"), whereas others have a natural distribution over multiple options (e.g., "the weighted coin flip was [$Heads/Tails$].") We argue that language model (LM) outputs should capture these natural distributions. Our work specifically tests whether LM output probabilities are calibrated to numeric information within their textual contexts. For example, if the context (the prompt) concerns two equally likely options (e.g., heads or tails for a fair coin), the LM output probabilities should also be equal. Likewise, in a context with nonuniformly likely events (e.g., rolling a pair with two dice) an LM should output proportionate probabilities. However, we find that even in simple settings, the best LMs (1) are poorly calibrated and (2) have systematic biases: artifacts like word identity, word order, and word frequency all impact calibration. For example, `gpt-4o-mini` often picks the first of two options presented in the prompt regardless of the options' implied likelihoods, whereas `Llama-3.1-8B` picks the second. Models do not allocate probability mass among valid options in a calibrated manner.

Language Model Probabilities are Not Calibrated in Numeric Contexts

Multi-document (MD) processing is crucial for LLMs to handle real-world tasks such as summarization and question-answering across large sets of documents. While LLMs have improved at processing long inputs, MD contexts still present unique difficulties, including management of inter-document dependencies, redundancy, and incoherent structures. To address this challenge, we introduce MDCure, a scalable and effective instruction data generation framework to enhance the MD capabilities of LLMs without the computational cost of pre-training or reliance on human-annotated data. MDCure generates high-quality synthetic MD instruction data over sets of articles via targeted prompts. We also introduce MDCureRM, a cost-effective, MD-specific reward model to score and filter generated data based on their training utility for MD settings. MDCure is compatible with open- and closed-source models in addition to policy optimization methods such as PPO, enabling even small open- source models to surpass proprietary LLMs as strong generators of high-quality MD instruction data without further data filtering. With MDCure, we fine-tune a wide variety of LLMs up to 70B parameters in size from the FlanT5, Qwen2, and LLAMA3.1 model families. Extensive evaluations on a wide range of MD and long-context benchmarks spanning various tasks and domains show MDCure consistently improves performance over pre-trained baselines and base models by up to 75.1%.

Downloads

Next from ACL 2025

SGIC: A Self-Guided Iterative Calibration Framework for RAG

Similar lecture

Tokenisation is NP-Complete

.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 ACL 2025

SGIC: A Self-Guided Iterative Calibration Framework for RAG

Similar lecture

Tokenisation is NP-Complete

Downloads