Aaron Mueller, Andrew Lee, Shruti Joshi, Ekdeep Singh Lubana, Dhanya Sridhar, Patrik Reizinger. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

A central goal of machine learning is generalization. While the No Free Lunch Theorem states that we cannot obtain theoretical guarantees for generalization without further assumptions, in practice we observe that simple models which explain the training data generalize best—a principle called Occam's razor. Despite the need for simple models, most current approaches in machine learning only minimize the training error, and at best indirectly promote simplicity through regularization or architecture design. Here, we draw a connection between Occam's razor and in-context learning—an emergent ability of certain sequence models like Transformers to learn at inference time from past observations in a sequence. In particular, we show that the next-token prediction loss used to train in-context learners is directly equivalent to a data compression technique called prequential coding, and that minimizing this loss amounts to jointly minimizing both the training error and the complexity of the model that was implicitly learned from context. Our theory and the empirical experiments we use to support it not only provide a normative account of in-context learning, but also elucidate the shortcomings of current in-context learning methods, suggesting ways in which they can be improved. We make our code available at https://github.com/3rdCore/PrequentialCode.

Large autoregressive models like Transformers can solve tasks through in-context learning (ICL) without learning new weights, suggesting avenues for efficiently solving new tasks. For many tasks, e.g., linear regression, the data factorizes: examples are independent given a task latent that generates the data, e.g., linear coefficients. While an optimal predictor leverages this factorization by inferring task latents, it is unclear if Transformers implicitly do so or instead exploit heuristics and statistical shortcuts through attention layers. In this paper, we systematically investigate the effect of explicitly inferring task latents by minimally modifying the Transformer architecture with a bottleneck to prevent shortcuts and incentivize structured solutions. We compare it against standard Transformers across various ICL tasks and find that contrary to intuition and recent works, there is little discernible difference between the two; biasing towards task-relevant latent variables does not lead to better out-of-distribution performance, in general. Curiously, we find that while the bottleneck effectively learns to extract latent task variables from context, downstream processing struggles to utilize them for robust prediction. Our study highlights the intrinsic limitations of Transformers in achieving structured ICL solutions that generalize, and shows that while inferring the right latents aids interpretability, it is not sufficient to alleviate this problem.

Language model activations entangle concepts that mediate their behavior, making it difficult to interpret these factors, which has implications for generalizability and robustness. We introduce an approach for disentangling these concepts without supervision. Existing methods for concept discovery often rely on external labels, contrastive prompts, or known causal structures, which limits their scalability and biases them toward predefined, easily annotatable features. In contrast, we propose a new unsupervised algorithm that identifies causal differentiating concepts—interpretable latent directions in LM activations that must be changed to elicit a different model behavior. These concepts are discovered using a constrained contrastive learning objective, guided by the insight that eliciting a target behavior requires only sparse changes to the underlying concepts. We formalize this notion and show that, under a particular assumption about the sparsity of these causal differentiating concepts, our method learns disentangled representations that align with human-interpretable factors influencing LM decisions. We empirically show the ability of our method to recover ground-truth causal factors in synthetic and semi-synthetic settings. Additionally, we illustrate the utility of our method through a case study on refusal behavior in language models. Our approach offers a scalable and interpretable lens into the internal workings of LMs, providing a principled foundation for interpreting language model behavior.

Synthetic control (SC) methods have been widely applied to estimate the causal effect of large-scale interventions, e.g., the state-wide effect of a change in policy. The idea of synthetic controls is to approximate one unit’s counterfactual outcomes using a weighted combination of some other units’ observed outcomes. The motivating question of this paper is: how does the SC strategy lead to valid causal inferences? We address this question by re-formulating the causal inference problem targeted by SC with a more fine-grained model, where we change the unit of analysis from “large units" (e.g., states) to “small units" (e.g., individuals in states). Under the re-formulation, we derive sufficient conditions for the non-parametric causal identification of the causal effect. We show that, in some settings, existing linear SC estimators are valid even when the data generating process is non-linear. We highlight two implications of the reformulation: 1) it clarifies where “linearity" comes from, and how it falls naturally out of the more fine-grained and flexible model; 2) it suggests new ways of using available data with SC methods for valid causal inference, in particular, new ways of selecting observations from which to estimate the counterfactual.

Abstract A fundamental goal of scientific research is to learn about causal relationships. However, despite its critical role in the life and social sciences, causality has not had the same importance in Natural Language Processing (NLP), which has traditionally placed more emphasis on predictive tasks. This distinction is beginning to fade, with an emerging area of interdisciplinary research at the convergence of causal inference and language processing. Still, research on causality in NLP remains scattered across domains without unified definitions, benchmark datasets and clear articulations of the challenges and opportunities in the application of causal inference to the textual domain, with its unique properties. In this survey, we consolidate research across academic areas and situate it in the broader NLP landscape. We introduce the statistical challenge of estimating causal effects with text, encompassing settings where text is used as an outcome, treatment, or to address confounding. In addition, we explore potential uses of causal inference to improve the robustness, fairness, and interpretability of NLP models. We thus provide a unified overview of causal inference for the NLP community.1

Abstract Researchers in the social sciences are often interested in the relationship between text and an outcome of interest, where the goal is to both uncover latent patterns in the text and predict outcomes for unseen texts. To this end, this paper develops the heterogeneous supervised topic model (HSTM), a probabilistic approach to text analysis and prediction. HSTMs posit a joint model of text and outcomes to find heterogeneous patterns that help with both text analysis and prediction. The main benefit of HSTMs is that they capture heterogeneity in the relationship between text and the outcome across latent topics. To fit HSTMs, we develop a variational inference algorithm based on the auto-encoding variational Bayes framework. We study the performance of HSTMs on eight datasets and find that they consistently outperform related methods, including fine-tuned black-box models. Finally, we apply HSTMs to analyze news articles labeled with pro- or anti-tone. We find evidence of differing language used to signal a pro- and anti-tone.

Instrumental variable methods provide a powerful approach to estimating causal effects in the presence of unobserved confounding. But a key challenge when applying them is the reliance on untestable "exclusion" assumptions that rule out any relationship between the instrument variable and the response that is not mediated by the treatment. In this paper, we show how to perform consistent IV estimation despite violations of the exclusion assumption. In particular, we show that when one has multiple candidate instruments, only a majority of these candidates—or, more generally, the modal candidate-response relationship—needs to be valid to estimate the causal effect. Our approach uses an estimate of the modal prediction from an ensemble of instrumental variable estimators. The technique is simple to apply and is "black-box" in the sense that it may be used with any instrumental variable estimator as long as the treatment effect is identified for each valid instrument independently. As such, it is compatible with recent machine-learning based estimators that allow for the estimation of conditional average treatment effects (CATE) on complex, high dimensional data. Experimentally, we achieve accurate estimates of conditional average treatment effects using an ensemble of deep network-based estimators, including on a challenging simulated Mendelian Randomization problem.

Reid Pryzant, Dallas Card, Dan Jurafsky, Victor Veitch, Dhanya Sridhar. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Recent mobile app technology lets people systematize the process of messaging their friends to urge them to vote. Prior to the most recent US midterm elections in 2018, the mobile app Outvote randomized an aspect of their system, hoping to unobtrusively assess the causal effect of their users’ messages on voter turnout. However, properly assessing this causal effect is hindered by multiple statistical challenges, including attenuation bias due to mismeasurement of subjects’ outcomes and low precision due to two-sided non-compliance with subjects’ assignments. We address these challenges, which are likely to impinge upon any study that seeks to randomize authentic friend-to-friend interactions, by tailoring the statistical analysis to make use of additional data about both users and subjects. Using meta-data of users’ in-app behavior, we reconstruct subjects’ positions in users’ queues. We use this information to refine the study population to more compliant subjects who were higher in the queues, and we do so in a systematic way which optimizes a proxy for the study’s power. To mitigate attenuation bias, we then use ancillary data of subjects’ matches to the voter rolls that lets us refine the study population to one with low rates of outcome mismeasurement. Our analysis reveals statistically significant treatment effects from friend-to-friend mobilization efforts ( 8.3, CI = (1.2, 15.3)) that are among the largest reported in the get-out-the-vote (GOTV) literature. While social pressure from friends has long been conjectured to play a role in effective GOTV treatments, the present study is among the first to assess these effects experimentally.

Does adding a theorem to a paper affect its chance of acceptance? Does labeling a post with the author’s gender affect the post popularity? This paper develops a method to estimate such causal effects from observational text data, adjusting for confounding features of the text such as the subject or writing quality. We assume that the text suffices for causal adjustment but that, in practice, it is prohibitively high-dimensional. To address this challenge, we develop causally sufficient embeddings, low- dimensional document representations that preserve sufficient information for causal identification and allow for efficient estimation of causal effects. Causally sufficient embeddings combine two ideas. The first is supervised dimensionality reduction: causal adjustment requires only the aspects of text that are predictive of both the treatment and outcome. The second is efficient language modeling: representations of text are designed to dispose of linguistically irrelevant information, and this information is also causally irrelevant. Our method adapts language models (specifically, word embeddings and topic models) to learn document embeddings that are able to predict both treatment and outcome. We study causally sufficient embeddings with semi-synthetic datasets and find that they improve causal estimation over related embedding methods. We illustrate the methods by answering the two motivating questions—the effect of a theorem on paper acceptance and the effect of a gender label on post popularity. Code and data available at github.com/vveitch/causal-text-embeddings-tf2.

Statistical methods applied to social media posts shed light on the dynamics of online dialogue. For example, users' wording choices predict their persuasiveness and users adopt the language patterns of other dialogue participants. In this paper, we estimate the causal effect of reply tones in debates on linguistic and sentiment changes in subsequent responses. The challenge for this estimation is that a reply's tone and subsequent responses are confounded by the users' ideologies on the debate topic and their emotions. To overcome this challenge, we learn representations of ideology using generative models of text. We study debates from 4Forums.com and compare annotated tones of replying such as emotional versus factual, or reasonable versus attacking. We show that our latent confounder representation reduces bias in ATE estimation. Our results suggest that factual and asserting tones affect dialogue and provide a methodology for estimating causal effects from text.

Complex causal networks underlie many real-world problems, from the regulatory interactions between genes to the environmental patterns used to understand climate change. Computational methods seek to infer these causal networks using observational data and domain knowledge. In this paper, we identify three key requirements for inferring the structure of causal networks for scientific discovery: (1) robustness to noise in observed measurements; (2) scalability to handle hundreds of variables; and (3) flexibility to encode domain knowledge and other structural constraints. We first formalize the problem of joint probabilistic causal structure discovery. We develop an approach using probabilistic soft logic (PSL) that exploits multiple statistical tests, supports efficient optimization over hundreds of variables, and can easily incorporate structural constraints, including imperfect domain knowledge. We compare our method against multiple well-studied approaches on biological and synthetic datasets, showing improvements of up to 20% in F1-score over the best performing baseline in realistic settings.

Alcoholism, also known as Alcohol Use Disorder (AUD), is a serious problem affecting millions of people worldwide. Recovery from AUD is known to be challenging and often leads to relapse at various points after enrolling in a rehabilitation program such as Alcoholics Anonymous (AA). In this work, we take a structured approach to understand recovery and relapse from AUD using social media data. To do so, we combine linguistic and psychological attributes of users with relational features that capture useful structure in the user interaction network. We evaluate our models on AA-attending users extracted from the Twitter social network and predict recovery at two different points---90 days and 1 year after the user joins AA, respectively. Our experiments reveal that our structured approach is helpful in predicting recovery in these users. We perform extensive quantitative analysis of different groups of features and dependencies among them. Our analysis sheds light on the role of each feature group and how they combine to predict recovery and relapse. Finally, we present a qualitative analysis of the different reasons behind users relapsing to AUD. Our models and analysis are helpful in making meaningful predictions in scenarios where only a subset of features are available and can potentially be helpful in identifying and preventing relapse early.

MOTIVATION: As concurrent use of multiple medications becomes ubiquitous among patients, it is crucial to characterize both adverse and synergistic interactions between drugs. Statistical methods for prediction of putative drug-drug interactions (DDIs) can guide in vitro testing and cut down significant cost and effort. With the abundance of experimental data characterizing drugs and their associated targets, such methods must effectively fuse multiple sources of information and perform inference over the network of drugs. RESULTS: We propose a probabilistic approach for jointly inferring unknown DDIs from a network of multiple drug-based similarities and known interactions. We use the highly scalable and easily extensible probabilistic programming framework Probabilistic Soft Logic We compare against two methods including a state-of-the-art DDI prediction system across three experiments and show best performing improvements of more than 50% in AUPR over both baselines. We find five novel interactions validated by external sources among the top-ranked predictions of our model. AVAILABILITY AND IMPLEMENTATION: Final versions of all datasets and implementations will be made publicly available. CONTACT: [email protected].

Dhanya Sridhar, James Foulds, Bert Huang, Lise Getoor, Marilyn Walker. Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2015.