Nils Holzenberger

Neel Guha, Julian Nyarko, Daniel E. Ho et al.

The advent of large language models (LLMs) and their adoption by the legal community has given rise to the question: what types of legal reasoning can LLMs perform? To enable greater study of this question, we present LegalBench: a collaboratively constructed legal reasoning benchmark consisting of 162 tasks covering six different types of legal reasoning. LegalBench was built through an interdisciplinary process, in which we collected tasks designed and hand-crafted by legal professionals. Because these subject matter experts took a leading role in construction, tasks either measure legal reasoning capabilities that are practically useful, or measure reasoning skills that lawyers find interesting. To enable cross-disciplinary conversations about LLMs in the law, we additionally show how popular legal frameworks for describing legal reasoning -- which distinguish between its many forms -- correspond to LegalBench tasks, thus giving lawyers and LLM developers a common vocabulary. This paper describes LegalBench, presents an empirical evaluation of 20 open-source and commercial LLMs, and illustrates the types of research explorations LegalBench enables.

Andrew Blair-Stanek, Nils Holzenberger, Benjamin Van Durme

Statutory reasoning is the task of reasoning with facts and statutes, which are rules written in natural language by a legislature. It is a basic legal skill. In this paper we explore the capabilities of the most capable GPT-3 model, text-davinci-003, on an established statutory-reasoning dataset called SARA. We consider a variety of approaches, including dynamic few-shot prompting, chain-of-thought prompting, and zero-shot prompting. While we achieve results with GPT-3 that are better than the previous best published results, we also identify several types of clear errors it makes. We investigate why these errors happen. We discover that GPT-3 has imperfect prior knowledge of the actual U.S. statutes on which SARA is based. More importantly, we create simple synthetic statutes, which GPT-3 is guaranteed not to have seen during training. We find GPT-3 performs poorly at answering straightforward questions about these simple synthetic statutes.

Guangyao Dou, William Jurayj, Nils Holzenberger et al.

Deontic reasoning is the task of answering questions by applying explicit rules and policies to case-specific facts, for example computing tax liability under a statute or determining the outcome of an immigration appeal. A key technical challenge for LLM-based deontic reasoning is that the relevant ruleset can be long and cross-referenced, so models may still fail to locate the rules needed for a particular reasoning step. We introduce Deontic Agentic Reasoning (DAR), an agentic reasoning setup in which the model interacts with the statutes on demand. We evaluate DAR under multiple harnesses on hard subsets of DeonticBench. Across these settings, we find that agentic harnesses can push the frontier on deontic reasoning tasks, but improvements are not uniform: weaker models often degrade on numerical tasks while consuming far more tokens.

Andrew Blair-Stanek, Nils Holzenberger, Benjamin Van Durme

We find that the best publicly available LLMs like GPT-4 and Claude currently perform poorly on basic legal text handling. This motivates the creation of a benchmark consisting of examples that lawyers and paralegals would expect LLMs to handle zero-shot, such as looking up the text at a line of a witness deposition or at a subsection of a contract. LLMs' poor performance on this benchmark casts into doubt their reliability as-is for legal practice. However, fine-tuning on our training set brings even a small model to near-perfect performance. This benchmark will be useful for fine-tuning LLMs for downstream legal tasks, as well as for tracking LLMs' reliability as-is for basic legal tasks.

Andrew Blair-Stanek, Nils Holzenberger, Benjamin Van Durme

The authors explain where OpenAI got the tax law example in its livestream demonstration of GPT-4, why GPT-4 got the wrong answer, and how it fails to reliably calculate taxes.

Nils Holzenberger, Yunmo Chen, Benjamin Van Durme

Information Extraction (IE) researchers are mapping tasks to Question Answering (QA) in order to leverage existing large QA resources, and thereby improve data efficiency. Especially in template extraction (TE), mapping an ontology to a set of questions can be more time-efficient than collecting labeled examples. We ask whether end users of TE systems can design these questions, and whether it is beneficial to involve an NLP practitioner in the process. We compare questions to other ways of phrasing natural language prompts for TE. We propose a novel model to perform TE with prompts, and find it benefits from questions over other styles of prompts, and that they do not require an NLP background to author.

Rajaa El Hamdani, Thomas Bonald, Fragkiskos Malliaros et al.

This paper investigates the factuality of large language models (LLMs) as knowledge bases in the legal domain, in a realistic usage scenario: we allow for acceptable variations in the answer, and let the model abstain from answering when uncertain. First, we design a dataset of diverse factual questions about case law and legislation. We then use the dataset to evaluate several LLMs under different evaluation methods, including exact, alias, and fuzzy matching. Our results show that the performance improves significantly under the alias and fuzzy matching methods. Further, we explore the impact of abstaining and in-context examples, finding that both strategies enhance precision. Finally, we demonstrate that additional pre-training on legal documents, as seen with SaulLM, further improves factual precision from 63% to 81%.

Samy Haffoudhi, Nikola Dobričić, Fabian Suchanek et al.

Entity linking is a key component of many downstream NLP systems, yet existing approaches are often tied to the specific target knowledge bases and domains, limiting their real world application. In this paper, we extend LELA, a modular and domain-agnostic LLM-based entity disambiguation method, into a practical Python library that integrates zero-shot Named Entity Recognition (NER) -thereby providing a complete end-toend pipeline for entity-linking in real-world usage. We provide experimental results validating LELA's performance and robustness across diverse entity linking settings. In our demo, users can play with the system on their own input texts.

Abe Bohan Hou, William Jurayj, Nils Holzenberger et al.

Large Language Models (LLMs) show promise as a writing aid for professionals performing legal analyses. However, LLMs can often hallucinate in this setting, in ways difficult to recognize by non-professionals and existing text evaluation metrics. In this work, we pose the question: when can machine-generated legal analysis be evaluated as acceptable? We introduce the neutral notion of gaps, as opposed to hallucinations in a strict erroneous sense, to refer to the difference between human-written and machine-generated legal analysis. Gaps do not always equate to invalid generation. Working with legal experts, we consider the CLERC generation task proposed in Hou et al. (2024b), leading to a taxonomy, a fine-grained detector for predicting gap categories, and an annotated dataset for automatic evaluation. Our best detector achieves 67% F1 score and 80% precision on the test set. Employing this detector as an automated metric on legal analysis generated by SOTA LLMs, we find around 80% contain hallucinations of different kinds.

Rajaa El Hamdani, Samy Haffoudhi, Nils Holzenberger et al.

Language models (LMs) encode substantial factual knowledge, but often produce answers judged as incorrect. We hypothesize that many of these answers are actually correct, but are expressed in alternative surface forms that are dismissed due to an overly strict evaluation, leading to an underestimation of models' parametric knowledge. We propose Retrieval-Constrained Decoding (RCD), a decoding strategy that restricts model outputs to unique surface forms. We introduce YAGO-QA, a dataset of 19,137 general knowledge questions. Evaluating open-source LMs from 135M to 70B parameters, we show that standard decoding undervalues their knowledge. For instance, Llama-3.1-70B scores only 32.3% F1 with vanilla decoding but 46.0% with RCD. Similarly, Llama-3.1-8B reaches 33.0% with RCD, outperforming the larger model under vanilla decoding. We publicly share the code and dataset at https://github.com/Rajjaa/disambiguated-LLM.

Abe Bohan Hou, Orion Weller, Guanghui Qin et al.

Legal professionals need to write analyses that rely on citations to relevant precedents, i.e., previous case decisions. Intelligent systems assisting legal professionals in writing such documents provide great benefits but are challenging to design. Such systems need to help locate, summarize, and reason over salient precedents in order to be useful. To enable systems for such tasks, we work with legal professionals to transform a large open-source legal corpus into a dataset supporting two important backbone tasks: information retrieval (IR) and retrieval-augmented generation (RAG). This dataset CLERC (Case Law Evaluation Retrieval Corpus), is constructed for training and evaluating models on their ability to (1) find corresponding citations for a given piece of legal analysis and to (2) compile the text of these citations (as well as previous context) into a cogent analysis that supports a reasoning goal. We benchmark state-of-the-art models on CLERC, showing that current approaches still struggle: GPT-4o generates analyses with the highest ROUGE F-scores but hallucinates the most, while zero-shot IR models only achieve 48.3% recall@1000.

Samy Haffoudhi, Fabian M. Suchanek, Nils Holzenberger

Entity linking (mapping ambiguous mentions in text to entities in a knowledge base) is a foundational step in tasks such as knowledge graph construction, question-answering, and information extraction. Our method, LELA, is a modular coarse-to-fine approach that leverages the capabilities of large language models (LLMs), and works with different target domains, knowledge bases and LLMs, without any fine-tuning phase. Our experiments across various entity linking settings show that LELA is highly competitive with fine-tuned approaches, and substantially outperforms the non-fine-tuned ones.

Avrile Floro, Tamara Dhorasoo, Soline Pellez et al.

Computational methods applied to legal scholarship hold the promise of analyzing law at scale. We start from a simple question: how often do courts implicitly apply statutory rules? This requires distinguishing legal reasoning from semantic similarity. We focus on implicit citation of the French Civil Code in first-instance court decisions and introduce a benchmark of 1,015 passage-article pairs annotated by three legal experts. We show that expert disagreement predicts model failures. Inter-annotator agreement is moderate ($κ$ = 0.33) with 43% of disagreements involving the boundary between factual description and legal reasoning. Our supervised ensemble achieves F1 = 0.70 (77% accuracy), but this figure conceals an asymmetry: 68% of false positives fall on the 33% of cases where the annotators disagreed. Despite these limits, reframing the task as top-k ranking and leveraging multi-model consensus yields 76% precision at k = 200 in an unsupervised setting. Moreover, the remaining false positives tend to surface legally ambiguous applications rather than obvious errors.

Guangyao Dou, Luis Brena, Akhil Deo et al.

Reasoning with complex, context-specific rules remains challenging for large language models (LLMs). In legal and policy settings, this manifests as deontic reasoning: reasoning about obligations, permissions, and prohibitions under explicit rules. While many recent benchmarks emphasize short-context mathematical reasoning, fewer focus on long-context, high-stakes deontic reasoning. To address this gap, we introduce DEONTICBENCH, a benchmark of 6,232 tasks across U.S. federal taxes, airline baggage policies, U.S. immigration administration, and U.S. state housing law. These tasks can be approached in multiple ways, including direct reasoning in language or with the aid of symbolic computation. Besides free-form chain-of-thought reasoning, DEONTICBENCH enables an optional solver-based workflow in which models translate statutes and case facts into executable Prolog, leading to formal problem interpretations and an explicit program trace. We release reference Prolog programs for all instances. Across frontier LLMs and coding models, best hard-subset performance reaches only 44.4% on SARA Numeric and 46.6 macro-F1 on Housing. We further study training with supervised fine-tuning and reinforcement learning for symbolic program generation. Although training improves Prolog generation quality, current RL methods still fail to solve these tasks reliably. Overall, DEONTICBENCH provides a benchmark for studying context-grounded rule reasoning in real-world domains under both symbolic and non-symbolic settings.

Peter Fratrič, Nils Holzenberger, David Restrepo Amariles

The legislative process is the backbone of a state built on solid institutions. Yet, due to the complexity of laws -- particularly tax law -- policies may lead to inequality and social tensions. In this study, we introduce a novel prototype system designed to address the issues of tax loopholes and tax avoidance. Our hybrid solution integrates a natural language interface with a domain-specific language tailored for planning. We demonstrate on a case study how tax loopholes and avoidance schemes can be exposed. We conclude that our prototype can help enhance social welfare by systematically identifying and addressing tax gaps stemming from loopholes.

Xinrui Zou, Ming Zhang, Nathaniel Weir et al.

Statutory reasoning refers to the application of legislative provisions to a series of case facts described in natural language. We re-frame statutory reasoning as an analogy task, where each instance of the analogy task involves a combination of two instances of statutory reasoning. This increases the dataset size by two orders of magnitude, and introduces an element of interpretability. We show that this task is roughly as difficult to Natural Language Processing models as the original task. Finally, we come back to statutory reasoning, solving it with a combination of a retrieval mechanism and analogy models, and showing some progress on prior comparable work.

William Jurayj, Nils Holzenberger, Benjamin Van Durme

According to the United States Internal Revenue Service, ''the average American spends $\$270$ and 13 hours filing their taxes''. Even beyond the U.S., tax filing requires complex reasoning, combining application of overlapping rules with numerical calculations. Because errors can incur costly penalties, any automated system must deliver high accuracy and auditability, making modern large language models (LLMs) poorly suited for this task. We propose an approach that integrates LLMs with a symbolic solver to calculate tax obligations. We evaluate variants of this system on the challenging StAtutory Reasoning Assessment (SARA) dataset, and include a novel method for estimating the cost of deploying such a system based on real-world penalties for tax errors. We further show how combining up-front translation of plain-text rules into formal logic programs, combined with intelligently retrieved exemplars for formal case representations, can dramatically improve performance on this task and reduce costs to well below real-world averages. Our results demonstrate the promise and economic feasibility of neuro-symbolic architectures for increasing equitable access to reliable tax assistance.

Andrew Blair-Stanek, Nils Holzenberger, Benjamin Van Durme

We investigate whether large language models can discover and analyze U.S. tax-minimization strategies. This real-world domain challenges even seasoned human experts, and progress can reduce tax revenue lost from well-advised, wealthy taxpayers. We evaluate the most advanced LLMs on their ability to (1) interpret and verify tax strategies, (2) fill in gaps in partially specified strategies, and (3) generate complete, end-to-end strategies from scratch. This domain should be of particular interest to the LLM reasoning community: unlike synthetic challenge problems or scientific reasoning tasks, U.S. tax law involves navigating hundreds of thousands of pages of statutes, case law, and administrative guidance, all updated regularly. Notably, LLM-based reasoning identified an entirely novel tax strategy, highlighting these models' potential to revolutionize tax agencies' fight against tax abuse.

Nils Holzenberger, Benjamin Van Durme

Statutory reasoning is the task of determining whether a legal statute, stated in natural language, applies to the text description of a case. Prior work introduced a resource that approached statutory reasoning as a monolithic textual entailment problem, with neural baselines performing nearly at-chance. To address this challenge, we decompose statutory reasoning into four types of language-understanding challenge problems, through the introduction of concepts and structure found in Prolog programs. Augmenting an existing benchmark, we provide annotations for the four tasks, and baselines for three of them. Models for statutory reasoning are shown to benefit from the additional structure, improving on prior baselines. Further, the decomposition into subtasks facilitates finer-grained model diagnostics and clearer incremental progress.

Noah Weber, Anton Belyy, Nils Holzenberger et al.

Event schemas are structured knowledge sources defining typical real-world scenarios (e.g., going to an airport). We present a framework for efficient human-in-the-loop construction of a schema library, based on a novel script induction system and a well-crafted interface that allows non-experts to "program" complex event structures. Associated with this work we release a schema library: a machine readable resource of 232 detailed event schemas, each of which describe a distinct typical scenario in terms of its relevant sub-event structure (what happens in the scenario), participants (who plays a role in the scenario), fine-grained typing of each participant, and the implied relational constraints between them. We make our schema library and the SchemaBlocks interface available online.

Nils Holzenberger, Andrew Blair-Stanek, Benjamin Van Durme

Legislation can be viewed as a body of prescriptive rules expressed in natural language. The application of legislation to facts of a case we refer to as statutory reasoning, where those facts are also expressed in natural language. Computational statutory reasoning is distinct from most existing work in machine reading, in that much of the information needed for deciding a case is declared exactly once (a law), while the information needed in much of machine reading tends to be learned through distributional language statistics. To investigate the performance of natural language understanding approaches on statutory reasoning, we introduce a dataset, together with a legal-domain text corpus. Straightforward application of machine reading models exhibits low out-of-the-box performance on our questions, whether or not they have been fine-tuned to the legal domain. We contrast this with a hand-constructed Prolog-based system, designed to fully solve the task. These experiments support a discussion of the challenges facing statutory reasoning moving forward, which we argue is an interesting real-world task that can motivate the development of models able to utilize prescriptive rules specified in natural language.

Nils Holzenberger, Raman Arora

Canonical correlation analysis (CCA) is a popular technique for learning representations that are maximally correlated across multiple views in data. In this paper, we extend the CCA based framework for learning a multiview mixture model. We show that the proposed model and a set of simple heuristics yield improvements over standard CCA, as measured in terms of performance on downstream tasks. Our experimental results show that our correlation-based objective meaningfully generalizes the CCA objective to a mixture of CCA models.

Nils Holzenberger, Shruti Palaskar, Pranava Madhyastha et al.

An increasing number of datasets contain multiple views, such as video, sound and automatic captions. A basic challenge in representation learning is how to leverage multiple views to learn better representations. This is further complicated by the existence of a latent alignment between views, such as between speech and its transcription, and by the multitude of choices for the learning objective. We explore an advanced, correlation-based representation learning method on a 4-way parallel, multimodal dataset, and assess the quality of the learned representations on retrieval-based tasks. We show that the proposed approach produces rich representations that capture most of the information shared across views. Our best models for speech and textual modalities achieve retrieval rates from 70.7% to 96.9% on open-domain, user-generated instructional videos. This shows it is possible to learn reliable representations across disparate, unaligned and noisy modalities, and encourages using the proposed approach on larger datasets.