Rule-Based Approach for Party-Based Sentiment Analysis in Legal Opinion Texts

Isanka Rajapaksha, Chanika Ruchini Mudalige, Dilini Karunarathna, Nisansa de Silva, Gathika Ratnayaka, Amal Shehan PereraarXiv 2020 · arXiv

Introduction

Legal opinion texts are characterized by their density, length, and structural complexity. For legal professionals, extracting specific viewpoints or the standing of various parties from vast document repositories is a labor-intensive task. Traditional sentiment analysis, which typically assigns a global polarity to a sentence, is insufficient for the legal domain. A single sentence in a court transcript often conveys divergent sentiments toward different entities—for instance, a "positive" action for the prosecution may simultaneously represent a "negative" outcome for the defendant. This research proposes a rule-based pipeline for Party-Based Sentiment Analysis, treating legal parties as "aspects" within an Aspect-Based Sentiment Analysis (ABSA) framework to capture these nuances.

Methodology

The proposed system utilizes a multi-stage NLP pipeline to decompose complex legal sentences and assign sentiment to specific parties.

The rule-based pipeline for party-based sentiment analysis.
  1. Co-reference Resolution: The system employs the Stanford Co-reference Resolution model to link pronouns (e.g., "he", "they") to their respective legal parties (e.g., "The Petitioner").
  2. Sentence Decomposition: Given the prevalence of long, subordinate-clause-heavy sentences in legal writing, a constituency parser identifies SBAR tags to split complex sentences into manageable sub-sentences.
  3. Phrase Extraction: Each sub-sentence is further parsed into Noun Phrases (NP) and Verb Phrases (VP).
  4. Sentiment Annotation: A specialized phrase-level sentiment annotator, specifically trained on legal corpora via transfer learning, classifies phrases as "negative" or "non-negative."
  5. Sentiment Allocation (The Principal Rule): The core logic of the approach dictates that the sentiment of a Verb Phrase is assigned to its object, while the inverse sentiment is assigned to the subject (the Noun Phrase).
  6. Aggregation: Sentiment scores across various phrases are averaged to determine the final polarity for the primary parties involved (e.g., Petitioner vs. Defendant).

Key Insights

The methodology introduces two significant conceptual shifts for legal NLP. First, it frames legal parties as "aspects," allowing the system to handle the multi-entity nature of judicial proceedings. Second, it leverages a "zero-sum" heuristic: in legal contexts, a negative action performed by a subject against an object often implies a specific framing for the subject that differs from the object's experience. By using structural splitting via SBAR tags, the authors acknowledge that legal language requires a hierarchical rather than "flat" analysis.

Results and Discussion

The authors provide a qualitative proof-of-concept using the case Lee v. United States. The system successfully analyzed sentences regarding federal officials receiving tips about illegal activity, correctly identifying positive sentiment for the government (the officials) and negative sentiment for the individual (Lee). This demonstrates that phrase-level sentiment, when processed through a structured rule-based pipeline, can be effectively aggregated to produce accurate party-based sentiment profiles.

Conclusion and Future Work

While the rule-based approach shows promise, several areas for development remain. The current binary classification ("negative" vs. "non-negative") may lack the granularity needed for complex legal interpretations. Furthermore, the rigidity of rule-based systems may struggle with linguistic nuances such as sarcasm or idiosyncratic judicial styles. Future research should focus on quantitative validation against large-scale human-annotated datasets and expanding the system to handle cases with multiple intervenors or complex third-party relationships.

Summary updated

Abstract

A document which elaborates opinions and arguments related to the previous court cases is known as a legal opinion text. Lawyers and legal officials have to spend considerable effort and time to obtain the required information manually from those documents when dealing with new legal cases. Hence, it provides much convenience to those individuals if there is a way to automate the process of extracting information from legal opinion texts. Party-based sentiment analysis will play a key role in the automation system by identifying opinion values with respect to each legal parties in legal texts.
BibTeX
@misc{rajapaksha2020rule,
  author = {Isanka Rajapaksha and Chanika Ruchini Mudalige and Dilini Karunarathna and Nisansa de Silva and Gathika Ratnayaka and Amal Shehan Perera},
  title = {{Rule-Based Approach for Party-Based Sentiment Analysis in Legal Opinion Texts}},
  year = {2020},
  eprint = {2011.05675},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2011.05675},
}
Figure