Bryan Li

Bryan Li, William Walden, Yu Hou et al.

Evaluation of long-form, citation-backed reports has lately received significant attention due to the wide-scale adoption of retrieval-augmented generation (RAG) systems. Core to many evaluation frameworks is the use of atomic facts, or nuggets, to assess a report's coverage of query-relevant information attested in the underlying collection. While nuggets have traditionally been represented as short statements, recent work has used question-answer (QA) representations, enabling fine-grained evaluations that decouple the information need (i.e. the question) from the potentially diverse content that satisfies it (i.e. its answers). A persistent challenge for nugget-based evaluation is the need to manually curate sets of nuggets for each topic in a test collection -- a laborious process that scales poorly to novel information needs. This challenge is acute in cross-lingual settings, where information is found in multilingual source documents. Accordingly, we introduce DoGMaTiQ, a pipeline for generating high-quality QA-based nugget sets in three stages: (1) document-grounded nugget generation, (2) paraphrase clustering, and (3) nugget subselection based on principled quality criteria. We integrate DoGMaTiQ nuggets with AutoArgue -- a recent nugget-based evaluation framework -- to enable fully automatic evaluation of generated reports. We conduct extensive experiments on two cross-lingual TREC shared tasks, NeuCLIR and RAGTIME, showing strong rank correlations with both human-in-the-loop and fully manual judgments. Finally, detailed analysis of our pipeline reveals that a strong LLM nugget generator is key, and that the system rankings induced by DoGMaTiQ are robust to outlier systems. We facilitate future research in report evaluation by publicly releasing our code and artifacts at https://github.com/manestay/dogmatiq.

Bryan Li, Chris Callison-Burch

Existing question answering (QA) systems owe much of their success to large, high-quality training data. Such annotation efforts are costly, and the difficulty compounds in the cross-lingual setting. Therefore, prior cross-lingual QA work has focused on releasing evaluation datasets, and then applying zero-shot methods as baselines. This work proposes a synthetic data generation method for cross-lingual QA which leverages indirect supervision from existing parallel corpora. Our method termed PAXQA (Projecting annotations for cross-lingual (x) QA) decomposes cross-lingual QA into two stages. First, we apply a question generation (QG) model to the English side. Second, we apply annotation projection to translate both the questions and answers. To better translate questions, we propose a novel use of lexically-constrained machine translation, in which constrained entities are extracted from the parallel bitexts. We apply PAXQA to generate cross-lingual QA examples in 4 languages (662K examples total), and perform human evaluation on a subset to create validation and test splits. We then show that models fine-tuned on these datasets outperform prior synthetic data generation models over several extractive QA datasets. The largest performance gains are for directions with non-English questions and English contexts. Ablation studies show that our dataset generation method is relatively robust to noise from automatic word alignments, showing the sufficient quality of our generations. To facilitate follow-up work, we release our code and datasets at https://github.com/manestay/paxqa .

Bryan Li, Mohammad Sadegh Rasooli, Ajay Patel et al.

We propose a two-stage approach for training a single NMT model to translate unseen languages both to and from English. For the first stage, we initialize an encoder-decoder model to pretrained XLM-R and RoBERTa weights, then perform multilingual fine-tuning on parallel data in 40 languages to English. We find this model can generalize to zero-shot translations on unseen languages. For the second stage, we leverage this generalization ability to generate synthetic parallel data from monolingual datasets, then bidirectionally train with successive rounds of back-translation. Our approach, which we EcXTra (English-centric Crosslingual (X) Transfer), is conceptually simple, only using a standard cross-entropy objective throughout. It is also data-driven, sequentially leveraging auxiliary parallel data and monolingual data. We evaluate unsupervised NMT results for 7 low-resource languages, and find that each round of back-translation training further refines bidirectional performance. Our final single EcXTra-trained model achieves competitive translation performance in all translation directions, notably establishing a new state-of-the-art for English-to-Kazakh (22.9 > 10.4 BLEU). Our code is available at https://github.com/manestay/EcXTra .

Divyansh Agarwal, Alexander R. Fabbri, Simeng Han et al. · salesforce

This paper introduces the shared task of summarizing documents in several creative domains, namely literary texts, movie scripts, and television scripts. Summarizing these creative documents requires making complex literary interpretations, as well as understanding non-trivial temporal dependencies in texts containing varied styles of plot development and narrative structure. This poses unique challenges and is yet underexplored for text summarization systems. In this shared task, we introduce four sub-tasks and their corresponding datasets, focusing on summarizing books, movie scripts, primetime television scripts, and daytime soap opera scripts. We detail the process of curating these datasets for the task, as well as the metrics used for the evaluation of the submissions. As part of the CREATIVESUMM workshop at COLING 2022, the shared task attracted 18 submissions in total. We discuss the submissions and the baselines for each sub-task in this paper, along with directions for facilitating future work in the field.

Ajay Patel, Bryan Li, Mohammad Sadegh Rasooli et al.

Large language models such as GPT-3 (Brown et al., 2020) can perform arbitrary tasks without undergoing fine-tuning after being prompted with only a few labeled examples. An arbitrary task can be reformulated as a natural language prompt, and a language model can be asked to generate the completion, indirectly performing the task in a paradigm known as prompt-based learning. To date, emergent prompt-based learning capabilities have mainly been demonstrated for unidirectional language models. However, bidirectional language models pre-trained on denoising objectives such as masked language modeling produce stronger learned representations for transfer learning. This motivates the possibility of prompting bidirectional models, but their pre-training objectives have made them largely incompatible with the existing prompting paradigm. We present SAP (Sequential Autoregressive Prompting), a technique that enables the prompting of bidirectional models. Utilizing the machine translation task as a case study, we prompt the bidirectional mT5 model (Xue et al., 2021) with SAP and demonstrate its few-shot and zero-shot translations outperform the few-shot translations of unidirectional models like GPT-3 and XGLM (Lin et al., 2021), despite mT5's approximately 50% fewer parameters. We further show SAP is effective on question answering and summarization. For the first time, our results demonstrate prompt-based learning is an emergent property of a broader class of language models, rather than only unidirectional models.

Alyssa Hwang, Bryan Li, Zhaoyi Hou et al.

With their remarkably improved text generation and prompting capabilities, large language models can adapt existing written information into forms that are easier to use and understand. In our work, we focus on recipes as an example of complex, diverse, and widely used instructions. We develop a prompt grounded in the original recipe and ingredients list that breaks recipes down into simpler steps. We apply this prompt to recipes from various world cuisines, and experiment with several large language models (LLMs), finding best results with GPT-3.5. We also contribute an Amazon Mechanical Turk task that is carefully designed to reduce fatigue while collecting human judgment of the quality of recipe revisions. We find that annotators usually prefer the revision over the original, demonstrating a promising application of LLMs in serving as digital sous chefs for recipes and beyond. We release our prompt, code, and MTurk template for public use.

Bryan Li, Aleksey Panasyuk, Chris Callison-Burch

We study how differences in persuasive language across Wikipedia articles, written in either English and Russian, can uncover each culture's distinct perspective on different subjects. We develop a large language model (LLM) powered system to identify instances of persuasive language in multilingual texts. Instead of directly prompting LLMs to detect persuasion, which is subjective and difficult, we propose to reframe the task to instead ask high-level questions (HLQs) which capture different persuasive aspects. Importantly, these HLQs are authored by LLMs themselves. LLMs over-generate a large set of HLQs, which are subsequently filtered to a small set aligned with human labels for the original task. We then apply our approach to a large-scale, bilingual dataset of Wikipedia articles (88K total), using a two-stage identify-then-extract prompting strategy to find instances of persuasion. We quantify the amount of persuasion per article, and explore the differences in persuasion through several experiments on the paired articles. Notably, we generate rankings of articles by persuasion in both languages. These rankings match our intuitions on the culturally-salient subjects; Russian Wikipedia highlights subjects on Ukraine, while English Wikipedia highlights the Middle East. Grouping subjects into larger topics, we find politically-related events contain more persuasion than others. We further demonstrate that HLQs obtain similar performance when posed in either English or Russian. Our methodology enables cross-lingual, cross-cultural understanding at scale, and we release our code, prompts, and data.

Bowen Jiang, Zhuoqun Hao, Young-Min Cho et al.

Large Language Models (LLMs) have emerged as personalized assistants for users across a wide range of tasks -- from offering writing support to delivering tailored recommendations or consultations. Over time, the interaction history between a user and an LLM can provide extensive information about an individual's traits and preferences. However, open questions remain on how well LLMs today can effectively leverage such history to (1) internalize the user's inherent traits and preferences, (2) track how the user profiling and preferences evolve over time, and (3) generate personalized responses accordingly in new scenarios. In this work, we introduce the PERSONAMEM benchmark. PERSONAMEM features curated user profiles with over 180 simulated user-LLM interaction histories, each containing up to 60 sessions of multi-turn conversations across 15 real-world tasks that require personalization. Given an in-situ user query, i.e. query issued by the user from the first-person perspective, we evaluate LLM chatbots' ability to identify the most suitable response according to the current state of the user's profile. We observe that current LLMs still struggle to recognize the dynamic evolution in users' profiles over time through direct prompting approaches. As a consequence, LLMs often fail to deliver responses that align with users' current situations and preferences, with frontier models such as GPT-4.1, o4-mini, GPT-4.5, o1, or Gemini-2.0 achieving only around 50% overall accuracy, suggesting room for improvement. We hope that PERSONAMEM, along with the user profile and conversation simulation pipeline, can facilitate future research in the development of truly user-aware chatbots. Code and data are available at github.com/bowen-upenn/PersonaMem.

Bryan Li

The word alignment task, despite its prominence in the era of statistical machine translation (SMT), is niche and under-explored today. In this two-part tutorial, we argue for the continued relevance for word alignment. The first part provides a historical background to word alignment as a core component of the traditional SMT pipeline. We zero-in on GIZA++, an unsupervised, statistical word aligner with surprising longevity. Jumping forward to the era of neural machine translation (NMT), we show how insights from word alignment inspired the attention mechanism fundamental to present-day NMT. The second part shifts to a survey approach. We cover neural word aligners, showing the slow but steady progress towards surpassing GIZA++ performance. Finally, we cover the present-day applications of word alignment, from cross-lingual annotation projection, to improving translation.

William Walden, Marc Mason, Orion Weller et al.

Generation of long-form, citation-backed reports is a primary use case for retrieval augmented generation (RAG) systems. While open-source evaluation tools exist for various RAG tasks, ones tailored to report generation (RG) are lacking. Accordingly, we introduce Auto-ARGUE, a robust LLM-based implementation of the recently proposed ARGUE framework for RG evaluation. We present analysis of Auto-ARGUE on the RG pilot task from the TREC 2024 NeuCLIR track, showing good system-level correlations with human judgments. We further release a web app for visualization of Auto-ARGUE outputs.

Bryan Li, Samar Haider, Chris Callison-Burch

Do the Spratly Islands belong to China, the Philippines, or Vietnam? A pretrained large language model (LLM) may answer differently if asked in the languages of each claimant country: Chinese, Tagalog, or Vietnamese. This contrasts with a multilingual human, who would likely answer consistently. In this paper, we show that LLMs recall certain geographical knowledge inconsistently when queried in different languages -- a phenomenon we term geopolitical bias. As a targeted case study, we consider territorial disputes, an inherently controversial and multilingual task. We introduce BorderLines, a dataset of territorial disputes which covers 251 territories, each associated with a set of multiple-choice questions in the languages of each claimant country (49 languages in total). We also propose a suite of evaluation metrics to precisely quantify bias and consistency in responses across different languages. We then evaluate various multilingual LLMs on our dataset and metrics to probe their internal knowledge and use the proposed metrics to discover numerous inconsistencies in how these models respond in different languages. Finally, we explore several prompt modification strategies, aiming to either amplify or mitigate geopolitical bias, which highlights how brittle LLMs are and how they tailor their responses depending on cues from the interaction context. Our code and data are available at https://github.com/manestay/borderlines

Faisal Ladhak, Bryan Li, Yaser Al-Onaizan et al.

We present a new summarization task, generating summaries of novel chapters using summary/chapter pairs from online study guides. This is a harder task than the news summarization task, given the chapter length as well as the extreme paraphrasing and generalization found in the summaries. We focus on extractive summarization, which requires the creation of a gold-standard set of extractive summaries. We present a new metric for aligning reference summary sentences with chapter sentences to create gold extracts and also experiment with different alignment methods. Our experiments demonstrate significant improvement over prior alignment approaches for our task as shown through automatic metrics and a crowd-sourced pyramid analysis. We make our data collection scripts available at https://github.com/manestay/novel-chapter-dataset .

Bryan Li, Tamer Alkhouli, Daniele Bonadiman et al.

The development of large language models (LLM) has shown progress on reasoning, though studies have largely considered either English or simple reasoning tasks. To address this, we introduce a multilingual structured reasoning and explanation dataset, termed xSTREET, that covers four tasks across six languages. xSTREET exposes a gap in base LLM performance between English and non-English reasoning tasks. We then propose two methods to remedy this gap, building on the insight that LLMs trained on code are better reasoners. First, at training time, we augment a code dataset with multilingual comments using machine translation while keeping program code as-is. Second, at inference time, we bridge the gap between training and inference by employing a prompt structure that incorporates step-by-step code primitives to derive new facts and find a solution. Our methods show improved multilingual performance on xSTREET, most notably on the scientific commonsense reasoning subtask. Furthermore, the models show no regression on non-reasoning tasks, thus demonstrating our techniques maintain general-purpose abilities.

Bryan Li, Jiaming Luo, Eleftheria Briakou et al. · mit

While large language models (LLMs) have been increasingly adopted for machine translation (MT), their performance for specialist domains such as medicine and law remains an open challenge. Prior work has shown that LLMs can be domain-adapted at test-time by retrieving targeted few-shot demonstrations or terminologies for inclusion in the prompt. Meanwhile, for general-purpose LLM MT, recent studies have found some success in generating similarly useful domain knowledge from an LLM itself, prior to translation. Our work studies domain-adapted MT with LLMs through a careful prompting setup, finding that demonstrations consistently outperform terminology, and retrieval consistently outperforms generation. We find that generating demonstrations with weaker models can close the gap with larger model's zero-shot performance. Given the effectiveness of demonstrations, we perform detailed analyses to understand their value. We find that domain-specificity is particularly important, and that the popular multi-domain benchmark is testing adaptation to a particular writing style more so than to a specific domain.

Laura Dietz, Bryan Li, Eugene Yang et al.

RAG systems are increasingly evaluated and optimized using LLM judges, an approach that is rapidly becoming the dominant paradigm for system assessment. Nugget-based approaches in particular are now embedded not only in evaluation frameworks but also in the architectures of RAG systems themselves. While this integration can lead to genuine improvements, it also creates a risk of faulty measurements due to circularity. In this paper, we investigate this risk through comparative experiments with nugget-based RAG systems, including Ginger and Crucible, against strong baselines such as GPT-Researcher. By deliberately modifying Crucible to generate outputs optimized for an LLM judge, we show that near-perfect evaluation scores can be achieved when elements of the evaluation - such as prompt templates or gold nuggets - are leaked or can be predicted. Our results highlight the importance of blind evaluation settings and methodological diversity to guard against mistaking metric overfitting for genuine system progress.

Laura Dietz, Bryan Li, Gabrielle Liu et al.

RAGE systems integrate ideas from automatic evaluation (E) into Retrieval-augmented Generation (RAG). As one such example, we present Crucible, a Nugget-Augmented Generation System that preserves explicit citation provenance by constructing a bank of Q&A nuggets from retrieved documents and uses them to guide extraction, selection, and report generation. Reasoning on nuggets avoids repeated information through clear and interpretable Q&A semantics - instead of opaque cluster abstractions - while maintaining citation provenance throughout the entire generation process. Evaluated on the TREC NeuCLIR 2024 collection, our Crucible system substantially outperforms Ginger, a recent nugget-based RAG system, in nugget recall, density, and citation grounding.

Jai Sharma, Yifan Wang, Bryan Li

Understanding dependencies between variables is critical for interpretability and efficient generation in masked diffusion models (MDMs), yet these models primarily expose marginal conditional distributions and do not explicitly represent inter-variable dependence. We propose a neural framework for estimating pairwise conditional mutual information (MI) directly from the hidden states of a pretrained MDM, using ground-truth MI computed from the model's own conditional distributions for supervision. The resulting estimator captures the model's internal belief about dependency structure and predicts the full MI matrix in a single forward pass, enabling MI-guided parallel decoding by identifying conditionally independent subsets of variables. We evaluate our approach on Sudoku and protein sequence generation with ESM-C, where the MI maps recover known structural constraints and enable a 3-5x magnitude reduction in inference-time forward passes compared to sequential decoding, while preserving generative quality and outperforming entropy-based parallelization methods.

Gabrielle Kaili-May Liu, Bryan Li, Arman Cohan et al.

Real-world use cases often present RAG systems with complex queries for which relevant information is missing from the corpus or is incomplete. In these settings, RAG systems must be able to reject unanswerable, out-of-scope queries and identify failures of retrieval and multi-hop reasoning. Despite this, existing RAG benchmarks rarely reflect realistic task complexity for multi-hop or out-of-scope questions, which often can be cheated via disconnected reasoning (i.e., solved without genuine multi-hop inference) or require only simple factual recall. This limits the ability for such benchmarks to uncover limitations of existing RAG systems. To address this gap, we present the first pipeline for automatic, difficulty-controlled creation of un$\underline{c}$heatable, $\underline{r}$ealistic, $\underline{u}$nanswerable, and $\underline{m}$ulti-hop $\underline{q}$uerie$\underline{s}$ (CRUMQs), adaptable to any corpus and domain. We use our pipeline to create CRUMQs over two popular RAG datasets and demonstrate its effectiveness via benchmark experiments on leading retrieval-augmented LLMs. Results show that compared to prior RAG benchmarks, CRUMQs are highly challenging for RAG systems and achieve up to 81.0\% reduction in cheatability scores. More broadly, our pipeline offers a simple way to enhance benchmark difficulty and realism and drive development of more capable RAG systems.

Bryan Li, Sounak Bagchi, Zizhan Wang

The increasing integration of Large Language Models (LLMs) into society necessitates robust defenses against vulnerabilities from jailbreaking and adversarial prompts. This project proposes a recursive framework for enhancing the resistance of LLMs to manipulation through the use of prompt simplification techniques. By increasing the transparency of complex and confusing adversarial prompts, the proposed method enables more reliable detection and prevention of malicious inputs. Our findings attempt to address a critical problem in AI safety and security, providing a foundation for the development of systems able to distinguish harmless inputs from prompts containing malicious intent. As LLMs continue to be used in diverse applications, the importance of such safeguards will only grow.

Bryan Li, Lara J. Martin, Chris Callison-Burch

Transformers have been showing near-human performance on a variety of tasks, but they are not without their limitations. We discuss the issue of conflating results of transformers that are instructed to do multiple tasks simultaneously. In particular, we focus on the domain of commonsense reasoning within story prose, which we call contextual commonsense inference (CCI). We look at the GLUCOSE (Mostafazadeh et al. 2020) dataset and task for predicting implicit commonsense inferences between story sentences. Since the GLUCOSE task simultaneously generates sentences and predicts the CCI relation, there is a conflation in the results. Is the model really measuring CCI or is its ability to generate grammatical text carrying the results? In this paper, we introduce the task contextual commonsense inference in sentence selection ($\rm{C {\small IS}}^2$), a simplified task that avoids conflation by eliminating language generation altogether. Our findings emphasize the necessity of future work to disentangle language generation from the desired NLP tasks at hand.

Bryan Li, Xinyue Wang, Homayoon Beigi

We propose a system to develop a basic automatic speech recognizer(ASR) for Cantonese, a low-resource language, through transfer learning of Mandarin, a high-resource language. We take a time-delayed neural network trained on Mandarin, and perform weight transfer of several layers to a newly initialized model for Cantonese. We experiment with the number of layers transferred, their learning rates, and pretraining i-vectors. Key findings are that this approach allows for quicker training time with less data. We find that for every epoch, log-probability is smaller for transfer learning models compared to a Cantonese-only model. The transfer learning models show slight improvement in CER.