Alexander Spangher

Yufei Tian, Tenghao Huang, Miri Liu et al.

This paper investigates the capability of LLMs in storytelling, focusing on narrative development and plot progression. We introduce a novel computational framework to analyze narratives through three discourse-level aspects: i) story arcs, ii) turning points, and iii) affective dimensions, including arousal and valence. By leveraging expert and automatic annotations, we uncover significant discrepancies between the LLM- and human- written stories. While human-written stories are suspenseful, arousing, and diverse in narrative structures, LLM stories are homogeneously positive and lack tension. Next, we measure narrative reasoning skills as a precursor to generative capacities, concluding that most LLMs fall short of human abilities in discourse understanding. Finally, we show that explicit integration of aforementioned discourse features can enhance storytelling, as is demonstrated by over 40% improvement in neural storytelling in terms of diversity, suspense, and arousal.

Alexander Spangher, Xiang Ren, Jonathan May et al.

News article revision histories provide clues to narrative and factual evolution in news articles. To facilitate analysis of this evolution, we present the first publicly available dataset of news revision histories, NewsEdits. Our dataset is large-scale and multilingual; it contains 1.2 million articles with 4.6 million versions from over 22 English- and French-language newspaper sources based in three countries, spanning 15 years of coverage (2006-2021). We define article-level edit actions: Addition, Deletion, Edit and Refactor, and develop a high-accuracy extraction algorithm to identify these actions. To underscore the factual nature of many edit actions, we conduct analyses showing that added and deleted sentences are more likely to contain updating events, main content and quotes than unchanged sentences. Finally, to explore whether edit actions are predictable, we introduce three novel tasks aimed at predicting actions performed during version updates. We show that these tasks are possible for expert humans but are challenging for large NLP models. We hope this can spur research in narrative framing and help provide predictive tools for journalists chasing breaking news.

Guillaume Sanchez, Honglu Fan, Alexander Spangher et al.

Classifier-Free Guidance (CFG) has recently emerged in text-to-image generation as a lightweight technique to encourage prompt-adherence in generations. In this work, we demonstrate that CFG can be used broadly as an inference-time technique in pure language modeling. We show that CFG (1) improves the performance of Pythia, GPT-2 and LLaMA-family models across an array of tasks: Q\&A, reasoning, code generation, and machine translation, achieving SOTA on LAMBADA with LLaMA-7B over PaLM-540B; (2) brings improvements equivalent to a model with twice the parameter-count; (3) can stack alongside other inference-time methods like Chain-of-Thought and Self-Consistency, yielding further improvements in difficult tasks; (4) can be used to increase the faithfulness and coherence of assistants in challenging form-driven and content-driven prompts: in a human evaluation we show a 75\% preference for GPT4All using CFG over baseline.

Alexander Spangher, Xinyu Hua, Yao Ming et al.

While GPT-2 generates sentences that are remarkably human-like, longer documents can ramble and do not follow human-like writing structure. We study the problem of imposing structure on long-range text. We propose a novel controlled text generation task, sequentially controlled text generation, and identify a dataset, NewsDiscourse as a starting point for this task. We develop a sequential controlled text generation pipeline with generation and editing. We test different degrees of structural awareness and show that, in general, more structural awareness results in higher control-accuracy, grammaticality, coherency and topicality, approaching human-level writing performance.

Mirac Suzgun, Emily Shen, Federico Bianchi et al.

AI chatbots are rapidly shaping how people encounter the news, yet no prior study has systematically measured how accurately these systems, with their proprietary search integrations and retrieval-synthesis pipelines, handle emerging facts across languages and regions. We present a 14-day (February 9-22, 2026) evaluation of six AI chatbots (Gemini 3 Flash and Pro, Grok 4, Claude 4.5 Sonnet, GPT-5 and GPT-4o mini) on 2,100 factual questions derived from same-day BBC News reporting across six regional services (US & Canada, Arabic, Afrique, Hindi, Russian, Turkish). The best systems achieve over 90% multiple-choice accuracy on questions about events reported hours earlier. The same systems, however, lose 11-13% under free-response evaluation, and 16-17% across the cohort. We further characterize three failure patterns. First, every model achieves its lowest accuracy on Hindi (79% vs. 89-91% elsewhere) and citations indicate an Anglophone retrieval bias (e.g., models answering Hindi queries cite English Wikipedia more than any Hindi outlet). Second, retrieval, not reasoning, failures drive over 70% of all errors. When models retrieve a correct source, they often extract the correct answer; the problem is to land on the right source in the first place. Third, models achieving 88-96% accuracy on well-formed questions drop to 19-70% when questions contain subtle false premises, with the most vulnerable model accepting fabricated facts 64% of the time. We also identify a detection-accuracy paradox: the best false-premise detector ranks second in adversarial accuracy (abstention rate), while a weaker detector ranks first, showing that premise detection and answer recovery are partially independent capabilities. Overall, these suggest that high accuracy can mask systematic regional inequity, near-total dependence on retrieval infrastructure, and vulnerability to imperfect queries real users pose.

Alexander Spangher, Divya Choudhary

Developing and improving computational approaches to covering news can increase journalistic output and improve the way stories are covered. In this work we approach the problem of covering crime stories in Los Angeles. We present a machine-in-the-loop system that covers individual crimes by (1) learning the prototypical coverage archetypes from classical news articles on crime to learn their structure and (2) using output from the Los Angeles Police department to generate "lede paragraphs", first structural unit of crime-articles. We introduce a probabilistic graphical model for learning article structure and a rule-based system for generating ledes. We hope our work can lead to systems that use these components together to form the skeletons of news articles covering crime. This work was done for a class project in Jonathan May's Advanced Natural Language Processing Course, Fall, 2019.

Alexander Spangher, Emilio Ferrara, Ben Welsh et al.

Journalists must find stories in huge amounts of textual data (e.g. leaks, bills, press releases) as part of their jobs: determining when and why text becomes news can help us understand coverage patterns and help us build assistive tools. Yet, this is challenging because very few labelled links exist, language use between corpora is very different, and text may be covered for a variety of reasons. In this work we focus on news coverage of local public policy in the San Francisco Bay Area by the San Francisco Chronicle. First, we gather news articles, public policy documents and meeting recordings and link them using probabilistic relational modeling, which we show is a low-annotation linking methodology that outperforms other retrieval-based baselines. Second, we define a new task: newsworthiness prediction, to predict if a policy item will get covered. We show that different aspects of public policy discussion yield different newsworthiness signals. Finally we perform human evaluation with expert journalists and show our systems identify policies they consider newsworthy with 68% F1 and our coverage recommendations are helpful with an 84% win-rate.

Louis Bradshaw, Alexander Spangher, Stella Biderman et al.

While generative models for music composition are increasingly capable, their adoption by musicians is hindered by text-prompting, an asynchronous workflow disconnected from the embodied, responsive nature of instrumental performance. To address this, we introduce Aria-Duet, an interactive system facilitating a real-time musical duet between a human pianist and Aria, a state-of-the-art generative model, using a Yamaha Disklavier as a shared physical interface. The framework enables a turn-taking collaboration: the user performs, signals a handover, and the model generates a coherent continuation performed acoustically on the piano. Beyond describing the technical architecture enabling this low-latency interaction, we analyze the system's output from a musicological perspective, finding the model can maintain stylistic semantics and develop coherent phrasal ideas, demonstrating that such embodied systems can engage in musically sophisticated dialogue and open a promising new path for human-AI co-creation.

Alexander Spangher, James Youn, Matt DeButts et al.

Human writers plan, then write. For large language models (LLMs) to play a role in longer-form article generation, we must understand the planning steps humans make before writing. We explore one kind of planning, source-selection in news, as a case-study for evaluating plans in long-form generation. We ask: why do specific stories call for specific kinds of sources? We imagine a generative process for story writing where a source-selection schema is first selected by a journalist, and then sources are chosen based on categories in that schema. Learning the article's plan means predicting the schema initially chosen by the journalist. Working with professional journalists, we adapt five existing schemata and introduce three new ones to describe journalistic plans for the inclusion of sources in documents. Then, inspired by Bayesian latent-variable modeling, we develop metrics to select the most likely plan, or schema, underlying a story, which we use to compare schemata. We find that two schemata: stance and social affiliation best explain source plans in most documents. However, other schemata like textual entailment explain source plans in factually rich topics like "Science". Finally, we find we can predict the most suitable schema given just the article's headline with reasonable accuracy. We see this as an important case-study for human planning, and provides a framework and approach for evaluating other kinds of plans. We release a corpora, NewsSources, with annotations for 4M articles.

Michael Lu, Hyundong Justin Cho, Weiyan Shi et al.

Large Language Models (LLMs) have demonstrated impressive capabilities in generating coherent text but often struggle with grounding language and strategic dialogue. To address this gap, we focus on journalistic interviews, a domain rich in grounding communication and abundant in data. We curate a dataset of 40,000 two-person informational interviews from NPR and CNN, and reveal that LLMs are significantly less likely than human interviewers to use acknowledgements and to pivot to higher-level questions. Realizing that a fundamental deficit exists in multi-turn planning and strategic thinking, we develop a realistic simulated environment, incorporating source personas and persuasive elements, in order to facilitate the development of agents with longer-horizon rewards. Our experiments show that while source LLMs mimic human behavior in information sharing, interviewer LLMs struggle with recognizing when questions are answered and engaging persuasively, leading to suboptimal information extraction across model size and capability. These findings underscore the need for enhancing LLMs' strategic dialogue capabilities.

Louis Bradshaw, Honglu Fan, Alexander Spangher et al.

We study the capabilities of generative autoregressive transformer models trained on large amounts of symbolic solo-piano transcriptions. After first pretraining on approximately 60,000 hours of music, we use a comparatively smaller, high-quality subset, to finetune models to produce musical continuations, perform symbolic classification tasks, and produce general-purpose contrastive MIDI embeddings by adapting the SimCLR framework to symbolic music. When evaluating piano continuation coherence, our generative model outperforms leading symbolic generation techniques and remains competitive with proprietary audio generation models. On MIR classification benchmarks, frozen representations from our contrastive model achieve state-of-the-art results in linear probe experiments, while direct finetuning demonstrates the generalizability of pretrained representations, often requiring only a few hundred labeled examples to specialize to downstream tasks.

Ryan Lee, Alexander Spangher, Xuezhe Ma

A patent must be deemed novel and non-obvious in order to be granted by the US Patent Office (USPTO). If it is not, a US patent examiner will cite the prior work, or prior art, that invalidates the novelty and issue a non-final rejection. Predicting what claims of the invention should change given the prior art is an essential and crucial step in securing invention rights, yet has not been studied before as a learnable task. In this work we introduce the PatentEdits dataset, which contains 105K examples of successful revisions that overcome objections to novelty. We design algorithms to label edits sentence by sentence, then establish how well these edits can be predicted with large language models (LLMs). We demonstrate that evaluating textual entailment between cited references and draft sentences is especially effective in predicting which inventive claims remained unchanged or are novel in relation to prior art.

Ethan Hsu, Hong Meng Yam, Ines Bouissou et al.

A large portion of real-world data science tasks are complex and require multi-hop web-based interactions: finding appropriate data available on the internet, synthesizing real-time data of various modalities from different locations, and producing summarized analyses. Existing web benchmarks often focus on simplistic interactions, such as form submissions or e-commerce transactions, and often do not require diverse tool-using capabilities required for web based data science. Conversely, traditional data science benchmarks typically concentrate on static, often textually bound datasets and do not assess end-to-end workflows that encompass data acquisition, cleaning, analysis, and insight generation. In response, we introduce WebDS, the first end-to-end web-based data science benchmark. It comprises 870 web-based data science tasks across 29 diverse websites from structured government data portals to unstructured news media, challenging agents to perform complex, multi-step operations requiring the use of tools and heterogeneous data formats that better reflect the realities of modern data analytics. Evaluations of current SOTA LLM agents indicate significant performance gaps in accomplishing these tasks. For instance, Browser Use, which accomplishes 80% of tasks on Web Voyager, successfully completes only 15% of tasks in WebDS, which our analysis suggests is due to new failure modes like poor information grounding, repetitive behavior and shortcut-taking that agents performing WebDS' tasks display. By providing a more robust and realistic testing ground, WebDS sets the stage for significant advances in the development of practically useful LLM-based data science.

Ben Welsh, Naitian Zhou, Arda Kaz et al.

Information prioritization plays an important role in how humans perceive and understand the world. Homepage layouts serve as a tangible proxy for this prioritization. In this work, we present NewsHomepages, a large dataset of over 3,000 new website homepages (including local, national and topic-specific outlets) captured twice daily over a three-year period. We develop models to perform pairwise comparisons between news items to infer their relative significance. To illustrate that modeling organizational hierarchies has broader implications, we applied our models to rank-order a collection of local city council policies passed over a ten-year period in San Francisco, assessing their "newsworthiness". Our findings lay the groundwork for leveraging implicit organizational cues to deepen our understanding of information prioritization.

Alexander Spangher, Kung-Hsiang Huang, Hyundong Cho et al.

As events progress, news articles often update with new information: if we are not cautious, we risk propagating outdated facts. In this work, we hypothesize that linguistic features indicate factual fluidity, and that we can predict which facts in a news article will update using solely the text of a news article (i.e. not external resources like search engines). We test this hypothesis, first, by isolating fact-updates in large news revisions corpora. News articles may update for many reasons (e.g. factual, stylistic, narrative). We introduce the NewsEdits 2.0 taxonomy, an edit-intentions schema that separates fact updates from stylistic and narrative updates in news writing. We annotate over 9,200 pairs of sentence revisions and train high-scoring ensemble models to apply this schema. Then, taking a large dataset of silver-labeled pairs, we show that we can predict when facts will update in older article drafts with high precision. Finally, to demonstrate the usefulness of these findings, we construct a language model question asking (LLM-QA) abstention task. We wish the LLM to abstain from answering questions when information is likely to become outdated. Using our predictions, we show, LLM absention reaches near oracle levels of accuracy.

Alexander Spangher, Nanyun Peng, Jonathan May et al.

News articles are driven by the informational sources journalists use in reporting. Modeling when, how and why sources get used together in stories can help us better understand the information we consume and even help journalists with the task of producing it. In this work, we take steps toward this goal by constructing the largest and widest-ranging annotated dataset, to date, of informational sources used in news writing. We show that our dataset can be used to train high-performing models for information detection and source attribution. We further introduce a novel task, source prediction, to study the compositionality of sources in news articles. We show good performance on this task, which we argue is an important proof for narrative science exploring the internal structure of news articles and aiding in planning-based language generation, and an important step towards a source-recommendation system to aid journalists.

Alexander Spangher, Jonathan May

In this work, we create a web application to highlight the output of NLP models trained to parse and label discourse segments in law text. Our system is built primarily with journalists and legal interpreters in mind, and we focus on state-level law that uses U.S. Census population numbers to allocate resources and organize government. Our system exposes a corpus we collect of 6,000 state-level laws that pertain to the U.S. census, using 25 scrapers we built to crawl state law websites, which we release. We also build a novel, flexible annotation framework that can handle span-tagging and relation tagging on an arbitrary input text document and be embedded simply into any webpage. This framework allows journalists and researchers to add to our annotation database by correcting and tagging new data.

Alexander Spangher, Nanyun Peng, Jonathan May et al.

Journalists publish statements provided by people, or \textit{sources} to contextualize current events, help voters make informed decisions, and hold powerful individuals accountable. In this work, we construct an ontological labeling system for sources based on each source's \textit{affiliation} and \textit{role}. We build a probabilistic model to infer these attributes for named sources and to describe news articles as mixtures of these sources. Our model outperforms existing mixture modeling and co-clustering approaches and correctly infers source-type in 80\% of expert-evaluated trials. Such work can facilitate research in downstream tasks like opinion and argumentation mining, representing a first step towards machine-in-the-loop \textit{computational journalism} systems.

Alexander Spangher, Nanyun Peng, Jonathan May et al.

Journalists obtain "leads", or story ideas, by reading large corpora of government records: court cases, proposed bills, etc. However, only a small percentage of such records are interesting documents. We propose a model of "newsworthiness" aimed at surfacing interesting documents. We train models on automatically labeled corpora -- published newspaper articles -- to predict whether each article was a front-page article (i.e., \textbf{newsworthy}) or not (i.e., \textbf{less newsworthy}). We transfer these models to unlabeled corpora -- court cases, bills, city-council meeting minutes -- to rank documents in these corpora on "newsworthiness". A fine-tuned RoBERTa model achieves .93 AUC performance on heldout labeled documents, and .88 AUC on expert-validated unlabeled corpora. We provide interpretation and visualization for our models.

Alexander Spangher, Jonathan May

News article revision histories have the potential to give us novel insights across varied fields of linguistics and social sciences. In this work, we present, to our knowledge, the first publicly available dataset of news article revision histories, or NewsEdits. Our dataset is multilingual; it contains 1,278,804 articles with 4,609,430 versions from over 22 English- and French-language newspaper sources based in three countries. Across version pairs, we count 10.9 million added sentences; 8.9 million changed sentences and 6.8 million removed sentences. Within the changed sentences, we derive 72 million atomic edits. NewsEdits is, to our knowledge, the largest corpus of revision histories of any domain.

Alexander Spangher, Jonathan May, Sz-rung Shiang et al.

Small class-imbalanced datasets, common in many high-level semantic tasks like discourse analysis, present a particular challenge to current deep-learning architectures. In this work, we perform an extensive analysis on sentence-level classification approaches for the News Discourse dataset, one of the largest high-level semantic discourse datasets recently published. We show that a multitask approach can improve 7% Micro F1-score upon current state-of-the-art benchmarks, due in part to label corrections across tasks, which improve performance for underrepresented classes. We also offer a comparative review of additional techniques proposed to address resource-poor problems in NLP, and show that none of these approaches can improve classification accuracy in such a setting.

Berk Ustun, Alexander Spangher, Yang Liu

Machine learning models are increasingly used to automate decisions that affect humans - deciding who should receive a loan, a job interview, or a social service. In such applications, a person should have the ability to change the decision of a model. When a person is denied a loan by a credit score, for example, they should be able to alter its input variables in a way that guarantees approval. Otherwise, they will be denied the loan as long as the model is deployed. More importantly, they will lack the ability to influence a decision that affects their livelihood. In this paper, we frame these issues in terms of recourse, which we define as the ability of a person to change the decision of a model by altering actionable input variables (e.g., income vs. age or marital status). We present integer programming tools to ensure recourse in linear classification problems without interfering in model development. We demonstrate how our tools can inform stakeholders through experiments on credit scoring problems. Our results show that recourse can be significantly affected by standard practices in model development, and motivate the need to evaluate recourse in practice.

Alexander Spangher

Using nonparametric methods has been increasingly explored in Bayesian hierarchical modeling as a way to increase model flexibility. Although the field shows a lot of promise, inference in many models, including Hierachical Dirichlet Processes (HDP), remain prohibitively slow. One promising path forward is to exploit the submodularity inherent in Indian Buffet Process (IBP) to derive near-optimal solutions in polynomial time. In this work, I will present a brief tutorial on Bayesian nonparametric methods, especially as they are applied to topic modeling. I will show a comparison between different non-parametric models and the current state-of-the-art parametric model, Latent Dirichlet Allocation (LDA).