Jingwei Ni

Jingwei Ni, Zhijing Jin, Markus Freitag et al. · deepmind, eth-zurich

Human-translated text displays distinct features from naturally written text in the same language. This phenomena, known as translationese, has been argued to confound the machine translation (MT) evaluation. Yet, we find that existing work on translationese neglects some important factors and the conclusions are mostly correlational but not causal. In this work, we collect CausalMT, a dataset where the MT training data are also labeled with the human translation directions. We inspect two critical factors, the train-test direction match (whether the human translation directions in the training and test sets are aligned), and data-model direction match (whether the model learns in the same direction as the human translation direction in the dataset). We show that these two factors have a large causal effect on the MT performance, in addition to the test-model direction mismatch highlighted by existing work on the impact of translationese. In light of our findings, we provide a set of suggestions for MT training and evaluation. Our code and data are at https://github.com/EdisonNi-hku/CausalMT

Tao Chen, Gangwei Jiang, Pengyu Cheng et al.

Reward models (RMs) provide critical feedback signals for LLM post-training, notably in reinforced fine-tuning (RFT) and reinforcement learning (RL) pipelines. However, current reward evaluation relies on heterogeneous criteria such as rule-based verifiers, ground-truth references, procedural checklists, and complex rubrics, where a unified mechanism to integrate all types of evidence remains unexplored. To this end, we propose Skill Reward Model (Skill-RM), a unified framework that reformulates reward modeling as the execution of a reusable Reward-Evaluation Skill. By treating reward computation as a structured agentic task, Skill-RM provides a consistent interface to orchestrate heterogeneous resources, dynamically selecting and aggregating evidence tailored to the specific requirements of each input. This approach enables the reward model to move beyond static evaluation, ensuring consistency and transparency across diverse tasks. Extensive experiments on reward benchmarks and downstream applications, including best-of-N selection and reinforcement learning, demonstrate that Skill-RM consistently outperforms traditional judge baselines. Our findings suggest that Skill-RM not only provides a unified solution for reward modeling but also achieves superior performance through the strategic and dynamic orchestration of evidence. The code is at https://github.com/Qwen-Applications/Skill-RM.

Jingwei Ni, Yihao Liu, Xinpeng Liu et al. · eth-zurich

Equipping Large Language Model (LLM) agents with domain-specific skills is critical for tackling complex tasks. Yet, manual authoring creates a severe scalability bottleneck. Conversely, automated skill generation often yields fragile or fragmented results because it either relies on shallow parametric knowledge or sequentially overfits to non-generalizable trajectory-local lessons. To overcome this, we introduce Trace2Skill, a framework that mirrors how human experts author skills: by holistically analyzing broad execution experience before distilling it into a single, comprehensive guide. Instead of reacting sequentially to individual trajectories, Trace2Skill dispatches a parallel fleet of sub-agents to analyze a diverse pool of executions. It extracts trajectory-specific lessons and hierarchically consolidates them into a unified, conflict-free skill directory via inductive reasoning. Trace2Skill supports both deepening existing human-written skills and creating new ones from scratch. Experiments in challenging domains, such as spreadsheet, VisionQA and math reasoning, show that Trace2Skill significantly improves upon strong baselines, including Anthropic's official xlsx skills. Crucially, this trajectory-grounded evolution does not merely memorize task instances or model-specific quirks: evolved skills transfer across LLM scales and generalize to OOD settings. For example, skills evolved by Qwen3.5-35B on its own trajectories improved a Qwen3.5-122B agent by up to 57.65 absolute percentage points on WikiTableQuestions. Ultimately, our results demonstrate that complex agent experience can be packaged into highly transferable, declarative skills -- requiring no parameter updates, no external retrieval modules, and utilizing open-source models as small as 35B parameters.

Saeid Ashraf Vaghefi, Qian Wang, Veruska Muccione et al. · eth-zurich

Large Language Models (LLMs) have made significant progress in recent years, achieving remarkable results in question-answering tasks (QA). However, they still face two major challenges: hallucination and outdated information after the training phase. These challenges take center stage in critical domains like climate change, where obtaining accurate and up-to-date information from reliable sources in a limited time is essential and difficult. To overcome these barriers, one potential solution is to provide LLMs with access to external, scientifically accurate, and robust sources (long-term memory) to continuously update their knowledge and prevent the propagation of inaccurate, incorrect, or outdated information. In this study, we enhanced GPT-4 by integrating the information from the Sixth Assessment Report of the Intergovernmental (IPCC AR6), the most comprehensive, up-to-date, and reliable source in this domain. We present our conversational AI prototype, available at www.chatclimate.ai and demonstrate its ability to answer challenging questions accurately in three different QA scenarios: asking from 1) GPT-4, 2) chatClimate, and 3) hybrid chatClimate. The answers and their sources were evaluated by our team of IPCC authors, who used their expert knowledge to score the accuracy of the answers from 1 (very-low) to 5 (very-high). The evaluation showed that the hybrid chatClimate provided more accurate answers, highlighting the effectiveness of our solution. This approach can be easily scaled for chatbots in specific domains, enabling the delivery of reliable and accurate information.

Jingwei Ni, Julia Bingler, Chiara Colesanti-Senni et al. · eth-zurich

In the face of climate change, are companies really taking substantial steps toward more sustainable operations? A comprehensive answer lies in the dense, information-rich landscape of corporate sustainability reports. However, the sheer volume and complexity of these reports make human analysis very costly. Therefore, only a few entities worldwide have the resources to analyze these reports at scale, which leads to a lack of transparency in sustainability reporting. Empowering stakeholders with LLM-based automatic analysis tools can be a promising way to democratize sustainability report analysis. However, developing such tools is challenging due to (1) the hallucination of LLMs and (2) the inefficiency of bringing domain experts into the AI development loop. In this paper, we ChatReport, a novel LLM-based system to automate the analysis of corporate sustainability reports, addressing existing challenges by (1) making the answers traceable to reduce the harm of hallucination and (2) actively involving domain experts in the development loop. We make our methodology, annotated datasets, and generated analyses of 1015 reports publicly available.

Mubashara Akhtar, Anka Reuel, Prajna Soni et al. · meta-ai

Artificial Intelligence (AI) benchmarks play a central role in measuring progress in model development and guiding deployment decisions. However, many benchmarks quickly become saturated, meaning that they can no longer differentiate between the best-performing models, diminishing their long-term value. In this study, we analyze benchmark saturation across 60 Large Language Model (LLM) benchmarks selected from technical reports by major model developers. To identify factors driving saturation, we characterize benchmarks along 14 properties spanning task design, data construction, and evaluation format. We test five hypotheses examining how each property contributes to saturation rates. Our analysis reveals that nearly half of the benchmarks exhibit saturation, with rates increasing as benchmarks age. Notably, hiding test data (i.e., public vs. private) shows no protective effect, while expert-curated benchmarks resist saturation better than crowdsourced ones. Our findings highlight which design choices extend benchmark longevity and inform strategies for more durable evaluation.

Jingwei Ni, Julia Bingler, Chiara Colesanti-Senni et al. · eth-zurich

This paper introduces a novel approach to enhance Large Language Models (LLMs) with expert knowledge to automate the analysis of corporate sustainability reports by benchmarking them against the Task Force for Climate-Related Financial Disclosures (TCFD) recommendations. Corporate sustainability reports are crucial in assessing organizations' environmental and social risks and impacts. However, analyzing these reports' vast amounts of information makes human analysis often too costly. As a result, only a few entities worldwide have the resources to analyze these reports, which could lead to a lack of transparency. While AI-powered tools can automatically analyze the data, they are prone to inaccuracies as they lack domain-specific expertise. This paper introduces a novel approach to enhance LLMs with expert knowledge to automate the analysis of corporate sustainability reports. We christen our tool CHATREPORT, and apply it in a first use case to assess corporate climate risk disclosures following the TCFD recommendations. CHATREPORT results from collaborating with experts in climate science, finance, economic policy, and computer science, demonstrating how domain experts can be involved in developing AI tools. We make our prompt templates, generated data, and scores available to the public to encourage transparency.

Minjing Shi, Junling Wang, Jingwei Ni et al.

Identifying logical fallacies in everyday discourse is challenging for many people. This challenge is amplified in the era of Large Language Models (LLMs), where malicious agents can deploy fallacious arguments to disseminate misinformation at scale. In this work, we explore the potential of LLMs as part of the solution. We introduce LFTutor, an intelligent tutoring system which uses LLMs to tutor laypeople and help them learn about logical fallacies. LFTutor integrates intent-driven Socratic questioning and critical argumentation principles to actively engage learners to reflect on their reasoning. Through both automatic and human evaluations, we demonstrate that LFTutor significantly outperforms baseline LLMs lacking these pedagogical strategies. This work highlights the promise of combining LLMs with pedagogical scaffolding to foster critical thinking and argument literacy in the age of AI.

Tobias Schimanski, Imene Kolli, Yu Fan et al.

PDFs are the second-most used document type on the internet (after HTML). Yet, existing QA datasets commonly start from text sources or only address specific domains. In this paper, we present pdfQA, a multi-domain 2K human-annotated (real-pdfQA) and 2K synthetic dataset (syn-pdfQA) differentiating QA pairs in ten complexity dimensions (e.g., file type, source modality, source position, answer type). We apply and evaluate quality and difficulty filters on both datasets, obtaining valid and challenging QA pairs. We answer the questions with open-source LLMs, revealing existing challenges that correlate with our complexity dimensions. pdfQA presents a basis for end-to-end QA pipeline evaluation, testing diverse skill sets and local optimizations (e.g., in information retrieval or parsing).

Jingwei Ni, Ekaterina Fadeeva, Tianyi Wu et al.

Solving complex tasks usually requires LLMs to generate long multi-step reasoning chains. Previous work has shown that verifying the correctness of individual reasoning steps can further improve the performance and efficiency of LLMs on such tasks and enhance solution interpretability. However, existing verification approaches, such as Process Reward Models (PRMs), are either computationally expensive, limited to specific domains, or require large-scale human or model-generated annotations. Thus, we propose a lightweight alternative for step-level reasoning verification based on data-driven uncertainty scores. We train transformer-based uncertainty quantification heads (UHeads) that use the internal states of a frozen LLM to estimate the uncertainty of its reasoning steps during generation. The approach is fully automatic: target labels are generated either by another larger LLM (e.g., DeepSeek R1) or in a self-supervised manner by the original model itself. UHeads are both effective and lightweight, containing less than 10M parameters. Across multiple domains, including mathematics, planning, and general knowledge question answering, they match or even surpass the performance of PRMs that are up to 810x larger. Our findings suggest that the internal states of LLMs encode their uncertainty and can serve as reliable signals for reasoning verification, offering a promising direction toward scalable and generalizable introspective LLMs.

Yu Fan, Jingwei Ni, Jakob Merane et al. · eth-zurich

Long-form legal reasoning remains a key challenge for large language models (LLMs) in spite of recent advances in test-time scaling. To address this, we introduce \textsc{LEXam}, a novel benchmark derived from 340 law exams spanning 116 law school courses across a range of subjects and degree levels. The dataset comprises 4,886 law exam questions in English and German, including 2,841 long-form, open-ended questions and 2,045 multiple-choice questions. Besides reference answers, the open questions are also accompanied by explicit guidance outlining the expected legal reasoning approach such as issue spotting, rule recall, or rule application. Our evaluation on both open-ended and multiple-choice questions present significant challenges for current LLMs; in particular, they notably struggle with open questions that require structured, multi-step legal reasoning. Moreover, our results underscore the effectiveness of the dataset in differentiating between models with varying capabilities. Deploying an ensemble LLM-as-a-Judge paradigm with rigorous human expert validation, we demonstrate how model-generated reasoning steps can be evaluated consistently and accurately, closely aligning with human expert assessments. Our evaluation setup provides a scalable method to assess legal reasoning quality beyond simple accuracy metrics. We have open-sourced our code on https://github.com/LEXam-Benchmark/LEXam and released our data on https://huggingface.co/datasets/LEXam-Benchmark/LEXam. Project page: https://lexam-benchmark.github.io.

Jingwei Ni, Tobias Schimanski, Meihong Lin et al.

Retrieval Augmented Generation (RAG) is widely employed to ground responses to queries on domain-specific documents. But do RAG implementations leave out important information when answering queries that need an integrated analysis of information (e.g., Tell me good news in the stock market today.)? To address these concerns, RAG developers need to annotate information retrieval (IR) data for their domain of interest, which is challenging because (1) domain-specific queries usually need nuanced definitions of relevance beyond shallow semantic relevance; and (2) human or GPT-4 annotation is costly and cannot cover all (query, document) pairs (i.e., annotation selection bias), thus harming the effectiveness in evaluating IR recall. To address these challenges, we propose DIRAS (Domain-specific Information Retrieval Annotation with Scalability), a manual-annotation-free schema that fine-tunes open-sourced LLMs to consider nuanced relevance definition and annotate (partial) relevance labels with calibrated relevance scores. Extensive evaluation shows that DIRAS enables smaller (8B) LLMs to achieve GPT-4-level performance on annotating and ranking unseen (query, document) pairs, and is helpful for real-world RAG development. All code, LLM generations, and human annotations can be found in \url{https://github.com/EdisonNi-hku/DIRAS}.

Zhijing Jin, Julius von Kügelgen, Jingwei Ni et al.

The principle of independent causal mechanisms (ICM) states that generative processes of real world data consist of independent modules which do not influence or inform each other. While this idea has led to fruitful developments in the field of causal inference, it is not widely-known in the NLP community. In this work, we argue that the causal direction of the data collection process bears nontrivial implications that can explain a number of published NLP findings, such as differences in semi-supervised learning (SSL) and domain adaptation (DA) performance across different settings. We categorize common NLP tasks according to their causal direction and empirically assay the validity of the ICM principle for text data using minimum description length. We conduct an extensive meta-analysis of over 100 published SSL and 30 DA studies, and find that the results are consistent with our expectations based on causal insights. This work presents the first attempt to analyze the ICM principle in NLP, and provides constructive suggestions for future modeling choices. Code available at https://github.com/zhijing-jin/icm4nlp

Jingwei Ni, Minjing Shi, Dominik Stammbach et al. · eth-zurich

With the rise of generative AI, automated fact-checking methods to combat misinformation are becoming more and more important. However, factual claim detection, the first step in a fact-checking pipeline, suffers from two key issues that limit its scalability and generalizability: (1) inconsistency in definitions of the task and what a claim is, and (2) the high cost of manual annotation. To address (1), we review the definitions in related work and propose a unifying definition of factual claims that focuses on verifiability. To address (2), we introduce AFaCTA (Automatic Factual Claim deTection Annotator), a novel framework that assists in the annotation of factual claims with the help of large language models (LLMs). AFaCTA calibrates its annotation confidence with consistency along three predefined reasoning paths. Extensive evaluation and experiments in the domain of political speech reveal that AFaCTA can efficiently assist experts in annotating factual claims and training high-quality classifiers, and can work with or without expert supervision. Our analyses also result in PoliClaim, a comprehensive claim detection dataset spanning diverse political topics.

Tobias Schimanski, Jingwei Ni, Mathias Kraus et al. · eth-zurich

Advances towards more faithful and traceable answers of Large Language Models (LLMs) are crucial for various research and practical endeavors. One avenue in reaching this goal is basing the answers on reliable sources. However, this Evidence-Based QA has proven to work insufficiently with LLMs in terms of citing the correct sources (source quality) and truthfully representing the information within sources (answer attributability). In this work, we systematically investigate how to robustly fine-tune LLMs for better source quality and answer attributability. Specifically, we introduce a data generation pipeline with automated data quality filters, which can synthesize diversified high-quality training and testing data at scale. We further introduce four test sets to benchmark the robustness of fine-tuned specialist models. Extensive evaluation shows that fine-tuning on synthetic data improves performance on both in- and out-of-distribution. Furthermore, we show that data quality, which can be drastically improved by proposed quality filters, matters more than quantity in improving Evidence-Based QA.

Markus Leippold, Saeid Ashraf Vaghefi, Dominik Stammbach et al. · eth-zurich

This paper presents Climinator, a novel AI-based tool designed to automate the fact-checking of climate change claims. Utilizing an array of Large Language Models (LLMs) informed by authoritative sources like the IPCC reports and peer-reviewed scientific literature, Climinator employs an innovative Mediator-Advocate framework. This design allows Climinator to effectively synthesize varying scientific perspectives, leading to robust, evidence-based evaluations. Our model demonstrates remarkable accuracy when testing claims collected from Climate Feedback and Skeptical Science. Notably, when integrating an advocate with a climate science denial perspective in our framework, Climinator's iterative debate process reliably converges towards scientific consensus, underscoring its adeptness at reconciling diverse viewpoints into science-based, factual conclusions. While our research is subject to certain limitations and necessitates careful interpretation, our approach holds significant potential. We hope to stimulate further research and encourage exploring its applicability in other contexts, including political fact-checking and legal domains.

Jingwei Ni, Yu Fan, Vilém Zouhar et al. · eth-zurich

Variation in human annotation (i.e., disagreements) is common in NLP, often reflecting important information like task subjectivity and sample ambiguity. Modeling this variation is important for applications that are sensitive to such information. Although RLVR-style reasoning (Reinforcement Learning with Verifiable Rewards) has improved Large Language Model (LLM) performance on many tasks, it remains unclear whether such reasoning enables LLMs to capture informative variation in human annotation. In this work, we evaluate the influence of different reasoning settings on LLM disagreement modeling. We systematically evaluate each reasoning setting across model sizes, distribution expression methods, and steering methods, resulting in 60 experimental setups across 3 tasks. Surprisingly, our results show that RLVR-style reasoning degrades performance in disagreement modeling, while naive Chain-of-Thought (CoT) reasoning improves the performance of RLHF LLMs (RL from human feedback). These findings underscore the potential risk of replacing human annotators with reasoning LLMs, especially when disagreements are important.

Tobias Schimanski, Chiara Colesanti Senni, Glen Gostlow et al. · eth-zurich

Nature is an amorphous concept. Yet, it is essential for the planet's well-being to understand how the economy interacts with it. To address the growing demand for information on corporate nature disclosure, we provide datasets and classifiers to detect nature communication by companies. We ground our approach in the guidelines of the Taskforce on Nature-related Financial Disclosures (TNFD). Particularly, we focus on the specific dimensions of water, forest, and biodiversity. For each dimension, we create an expert-annotated dataset with 2,200 text samples and train classifier models. Furthermore, we show that nature communication is more prevalent in hotspot areas and directly effected industries like agriculture and utilities. Our approach is the first to respond to calls to assess corporate nature communication on a large scale.

Chenfei Xiong, Jingwei Ni, Yu Fan et al. · eth-zurich

We introduce Co-DETECT (Collaborative Discovery of Edge cases in TExt ClassificaTion), a novel mixed-initiative annotation framework that integrates human expertise with automatic annotation guided by large language models (LLMs). Co-DETECT starts with an initial, sketch-level codebook and dataset provided by a domain expert, then leverages the LLM to annotate the data and identify edge cases that are not well described by the initial codebook. Specifically, Co-DETECT flags challenging examples, induces high-level, generalizable descriptions of edge cases, and assists user in incorporating edge case handling rules to improve the codebook. This iterative process enables more effective handling of nuanced phenomena through compact, generalizable annotation rules. Extensive user study, qualitative and quantitative analyses prove the effectiveness of Co-DETECT.

Tianyi Wu, Jingwei Ni, Bryan Hooi et al. · eth-zurich

Instruction fine-tuning (IFT) can increase the informativeness of large language models (LLMs), but may reduce their truthfulness. This trade-off arises because IFT steers LLMs to generate responses containing long-tail knowledge that was not well covered during pre-training. As a result, models become more informative but less accurate when generalizing to unseen tasks. In this paper, we empirically demonstrate how unfamiliar knowledge in IFT datasets can negatively affect the truthfulness of LLMs, and we introduce two new IFT paradigms, $UNIT_{cut}$ and $UNIT_{ref}$, to address this issue. $UNIT_{cut}$ identifies and removes unfamiliar knowledge from IFT datasets to mitigate its impact on model truthfulness, whereas $UNIT_{ref}$ trains LLMs to recognize their uncertainty and explicitly indicate it at the end of their responses. Our experiments show that $UNIT_{cut}$ substantially improves LLM truthfulness, while $UNIT_{ref}$ maintains high informativeness and reduces hallucinations by distinguishing between confident and uncertain statements.

Yongan Yu, Xianda Du, Qingchen Hu et al.

Historical archives on weather events are collections of enduring primary source records that offer rich, untapped narratives of how societies have experienced and responded to extreme weather events. These qualitative accounts provide insights into societal vulnerability and resilience that are largely absent from meteorological records, making them valuable for climate scientists to understand societal responses. However, their vast scale, noisy digitized quality, and archaic language make it difficult to transform them into structured knowledge for climate research. To address this challenge, we introduce WeatherArchive-Bench, the first benchmark for evaluating retrieval-augmented generation (RAG) systems on historical weather archives. WeatherArchive-Bench comprises two tasks: WeatherArchive-Retrieval, which measures a system's ability to locate historically relevant passages from over one million archival news segments, and WeatherArchive-Assessment, which evaluates whether Large Language Models (LLMs) can classify societal vulnerability and resilience indicators from extreme weather narratives. Extensive experiments across sparse, dense, and re-ranking retrievers, as well as a diverse set of LLMs, reveal that dense retrievers often fail on historical terminology, while LLMs frequently misinterpret vulnerability and resilience concepts. These findings highlight key limitations in reasoning about complex societal indicators and provide insights for designing more robust climate-focused RAG systems from archival contexts. The constructed dataset and evaluation framework are publicly available at https://anonymous.4open.science/r/WeatherArchive-Bench/.

Alejandro Hernández-Cano, Alexander Hägele, Allen Hao Huang et al. · eth-zurich

We present Apertus, a fully open suite of large language models (LLMs) designed to address two systemic shortcomings in today's open model ecosystem: data compliance and multilingual representation. Unlike many prior models that release weights without reproducible data pipelines or regard for content-owner rights, Apertus models are pretrained exclusively on openly available data, retroactively respecting robots.txt exclusions and filtering for non-permissive, toxic, and personally identifiable content. To mitigate risks of memorization, we adopt the Goldfish objective during pretraining, strongly suppressing verbatim recall of data while retaining downstream task performance. The Apertus models also expand multilingual coverage, training on 15T tokens from over 1800 languages, with ~40% of pretraining data allocated to non-English content. Released at 8B and 70B scales, Apertus approaches state-of-the-art results among fully open models on multilingual benchmarks, rivalling or surpassing open-weight counterparts. Beyond model weights, we release all scientific artifacts from our development cycle with a permissive license, including data preparation scripts, checkpoints, evaluation suites, and training code, enabling transparent audit and extension.

Jingwei Ni, Zhijing Jin, Qian Wang et al.

Multi-task learning (MTL) aims at achieving a better model by leveraging data and knowledge from multiple tasks. However, MTL does not always work -- sometimes negative transfer occurs between tasks, especially when aggregating loosely related skills, leaving it an open question when MTL works. Previous studies show that MTL performance can be improved by algorithmic tricks. However, what tasks and skills should be included is less well explored. In this work, we conduct a case study in Financial NLP where multiple datasets exist for skills relevant to the domain, such as numeric reasoning and sentiment analysis. Due to the task difficulty and data scarcity in the Financial NLP domain, we explore when aggregating such diverse skills from multiple datasets with MTL can work. Our findings suggest that the key to MTL success lies in skill diversity, relatedness between tasks, and choice of aggregation size and shared capacity. Specifically, MTL works well when tasks are diverse but related, and when the size of the task aggregation and the shared capacity of the model are balanced to avoid overwhelming certain tasks.