Skip to content
Showcase/MAUD: Legal Merger Agreement Understanding
advancedtext

MAUD: Legal Merger Agreement Understanding

Legal document understanding for merger agreements. Annotators identify key clauses in merger agreements and answer structured questions about deal terms (termination fees, representations, conditions, etc.). Based on the MAUD dataset of expert-annotated legal NLP for merger agreement understanding.

About this dataset

MAUD (Merger Agreement Understanding Dataset) is an expert-annotated legal reading-comprehension dataset for question answering over merger-and-acquisition contracts. It was curated under the supervision of M&A lawyers and released by Wang et al. at EMNLP 2023. Given a clause from a merger agreement, the task is to answer a standardized legal question about the deal terms. It was built because merger agreements are long and technical and there was almost no expert-labeled data for training or evaluating legal NLP on them.

The source text comes from 152 English-language public merger agreements, the same deals used in the American Bar Association's 2021 Public Target Deal Points Study. Law students and experienced lawyers read each agreement, located the clauses relevant to each deal point, and selected the correct answer from a fixed option set for 92 reading-comprehension questions.

The dataset holds 47,457 annotations in total, made up of 8,226 clause-level deal-point annotations and 39,231 question-answer examples. The annotation work ran to more than 10,000 hours by law students and experienced lawyers, which is what makes the labels dependable in a domain where exact wording carries legal weight.

The Potato config below reproduces this task with two schemes: a span layer where annotators highlight the clause that answers the question and tag it with one of six deal categories (Deal Protection, Conditions to Closing, Material Adverse Effect, Termination, Representations & Warranties, Indemnification), and a radio layer for the structured answer (Yes, No, Partial/Qualified, Not Addressed, Ambiguous). It collects clause spans and their legal reading in a single pass.

Merger agreements
152 English-language public deals
Total annotations
47,457
Question-answer examples
39,231
Deal-point clause annotations
8,226
Reading-comprehension questions
92
Annotation effort
10,000+ expert hours
PERORGLOCPERORGLOCDATESelect text to annotate

Configuration Fileconfig.yaml

This Potato config reproduces the annotation task. Save it as config.yaml and run potato start config.yaml to try it.

yaml
# MAUD: Legal Merger Agreement Understanding
# Based on Wang et al., EMNLP 2023
# Paper: https://aclanthology.org/2023.emnlp-main.1009/
# Dataset: https://github.com/TheAtticusProject/maud
#
# Task: Legal document understanding for merger agreements
# Annotators identify key clauses and answer structured questions
# about deal terms in M&A agreements.
#
# Two-step annotation:
# 1. Highlight the relevant clause span(s) answering the question
# 2. Select the structured answer category for the question
#
# Categories cover: Deal Protection, Conditions to Closing,
# Material Adverse Effect, Termination, Representations & Warranties

annotation_task_name: "MAUD: Legal Merger Agreement Understanding"
task_dir: "."

data_files:
  - sample-data.json
item_properties:
  id_key: "id"
  text_key: "text"

output_annotation_dir: "annotation_output/"
output_annotation_format: "json"

annotation_schemes:
  # Step 1: Identify relevant clause spans
  - annotation_type: span
    name: clause_span
    description: "Highlight the clause or provision that answers the question"
    labels:
      - "Deal Protection"
      - "Conditions to Closing"
      - "Material Adverse Effect"
      - "Termination"
      - "Representations & Warranties"
      - "Indemnification"
    label_colors:
      "Deal Protection": "#3b82f6"
      "Conditions to Closing": "#22c55e"
      "Material Adverse Effect": "#ef4444"
      "Termination": "#f97316"
      "Representations & Warranties": "#8b5cf6"
      "Indemnification": "#eab308"
    tooltips:
      "Deal Protection": "Provisions protecting the deal (no-shop, matching rights, break-up fees)"
      "Conditions to Closing": "Conditions that must be satisfied before closing (regulatory approval, shareholder vote)"
      "Material Adverse Effect": "Clauses defining or referencing material adverse effects/changes"
      "Termination": "Provisions related to termination rights, fees, or conditions"
      "Representations & Warranties": "Statements of fact about the company (financial, legal, operational)"
      "Indemnification": "Provisions for indemnification, liability, or damage recovery"
    allow_overlapping: false

  # Step 2: Structured answer to the legal question
  - annotation_type: radio
    name: answer
    description: "Based on the clause, select the answer to the question shown above"
    labels:
      - "Yes"
      - "No"
      - "Partial / Qualified"
      - "Not Addressed"
      - "Ambiguous"
    keyboard_shortcuts:
      "Yes": "y"
      "No": "n"
      "Partial / Qualified": "p"
      "Not Addressed": "a"
      "Ambiguous": "b"
    tooltips:
      "Yes": "The agreement clearly and fully addresses the question in the affirmative"
      "No": "The agreement clearly addresses the question in the negative"
      "Partial / Qualified": "The agreement addresses the question but with qualifications, exceptions, or limitations"
      "Not Addressed": "The agreement does not contain a provision addressing this question"
      "Ambiguous": "The language is unclear or could be interpreted in multiple ways"

html_layout: |
  <div style="margin-bottom: 10px; padding: 10px; background: #f0f4ff; border-radius: 6px;">
    <strong>Category:</strong> {{category}}<br/>
    <strong>Question:</strong> {{question}}
  </div>
  <div style="padding: 10px; border: 1px solid #ddd; border-radius: 6px; line-height: 1.8;">
    {{text}}
  </div>

allow_all_users: true
instances_per_annotator: 50
annotation_per_instance: 2
allow_skip: true
skip_reason_required: false

Sample Datasample-data.json

json
[
  {
    "id": "maud_001",
    "text": "Section 5.3 No Solicitation. From the date of this Agreement until the earlier of the Effective Time or the termination of this Agreement, the Company shall not, and shall cause its Subsidiaries and its and their respective directors, officers, employees, investment bankers, attorneys, accountants and other advisors or representatives not to, directly or indirectly, (a) solicit, initiate, knowingly encourage or knowingly facilitate any Acquisition Proposal, (b) enter into, continue or otherwise participate in any discussions or negotiations regarding, or furnish to any person any information with respect to, any Acquisition Proposal, or (c) enter into any letter of intent, agreement in principle, acquisition agreement, merger agreement or similar agreement relating to any Acquisition Proposal.",
    "question": "Does the agreement contain a no-shop provision that prohibits the target from soliciting competing acquisition proposals?",
    "category": "Deal Protection"
  },
  {
    "id": "maud_002",
    "text": "Section 7.1 Conditions to Each Party's Obligations. The respective obligation of each party to effect the Merger shall be subject to the satisfaction or waiver at or prior to the Effective Time of the following conditions: (a) Stockholder Approval. The Company Stockholder Approval shall have been obtained. (b) Regulatory Approvals. All waiting periods applicable to the consummation of the Merger under the HSR Act shall have expired or been terminated, and all consents required under applicable Antitrust Laws shall have been obtained. (c) No Injunctions or Restraints. No Governmental Entity shall have enacted, issued, promulgated, enforced or entered any Law or Order that is then in effect and that has the effect of making the Merger illegal or otherwise restraining or prohibiting consummation of the Merger.",
    "question": "Is shareholder approval required as a condition to closing?",
    "category": "Conditions to Closing"
  }
]

// ... and 8 more items

Get This Design

View on GitHub

Clone or download from the repository

Quick start:

git clone https://github.com/davidjurgens/potato-showcase.git
cd potato-showcase/text/domain-specific/maud-legal-merger-qa
potato start config.yaml

Dataset & paper

Wang et al., EMNLP 2023

Official dataset ↗Read the paper ↗

Citation (BibTeX)

bibtex
@inproceedings{wang-etal-2023-maud,
    title = "{MAUD}: An Expert-Annotated Legal {NLP} Dataset for Merger Agreement Understanding",
    author = "Wang, Steven and Scardigli, Antoine and Tang, Leonard and Chen, Wei and Levkin, Dmitry and Chen, Anya and Ball, Spencer and Woodside, Thomas and Zhang, Oliver and Hendrycks, Dan",
    booktitle = "Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing",
    month = dec,
    year = "2023",
    address = "Singapore",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.emnlp-main.1019/",
    doi = "10.18653/v1/2023.emnlp-main.1019",
    pages = "16369--16382"
}

Details

Annotation Types

spanradio

Domain

NLPLegal

Use Cases

Legal Document UnderstandingClause IdentificationQuestion Answering

Tags

legalmerger-agreementclause-extractionmaudemnlp2023contract-analysis

Found an issue or want to improve this design?

Open an Issue

Related Designs

LegalEval - Legal Document Analysis

Relevance classification and named entity recognition for legal documents, identifying key entities such as courts, petitioners, respondents, judges, dates, statutes, and precedents. Based on SemEval-2023 Task 6 (Modi et al.).

radiospan

Aspect-Based Sentiment Analysis

Identification of aspect terms in review text with sentiment polarity classification for each aspect. Based on SemEval-2016 Task 5 (ABSA).

spanradio

Causal Medical Claim Detection and PICO Extraction

Detection of causal claims in medical texts and extraction of PICO (Population, Intervention, Comparator, Outcome) elements. Based on SemEval-2023 Task 8 (Khetan et al.).

spanradio