Shift-of-Perspective Identification Within Legal Cases

Gathika Ratnayaka, Thejan Rupasinghe, Nisansa de Silva, Viraj Salaka Gamage, Menuka Warushavithana, Amal Shehan PereraarXiv 2019 · arXiv

Introduction

Legal professionals spend significant time analyzing court opinions to extract core arguments, evidence, and counter-arguments. Identifying when a judge or lawyer pivots from a factual statement to a dissenting viewpoint is a critical yet challenging task for automated systems. In "Shift-of-Perspective Identification within Legal Cases," Ratnayaka et al. propose a multi-pronged NLP approach to detect these "Shift-in-View" relationships—instances where two sentences discuss the same topic but present differing perspectives.

A primary hurdle in legal discourse analysis is that standard machine learning models often fail to capture the nuance of legal jargon. Existing tools for Discourse Relation analysis frequently group sentence pairs under broad categories like "Elaboration." While technically accurate in some contexts, this classification obscures whether the elaboration supports or contradicts the preceding statement.

Confusion matrix showing the misclassification of Shift-in-View as Elaboration.

As shown in the confusion matrix above, baseline models struggle significantly with this distinction, often misclassifying 100% of perspective shifts as simple elaborations. This "gap" prevents effective automation of tasks such as counter-argument detection or party stance identification.

Proposed Methodology

The researchers developed a pipeline centered on a Sentence Relationship Identifier (SRI) to filter for topic-related sentences. Once a pair is deemed related, three specialized detection strategies are applied:

  1. Verb Relationship Analysis: This component proved most effective. It extracts and lemmatizes verbs, calculating semantic similarity via the Lin similarity score. It specifically identifies negation and polarities in adverbial modifiers.
  2. Sentiment-based Approach: Utilizing a sentiment annotator fine-tuned for the legal domain, the system identifies sub-clauses with opposing sentiment polarities regarding the same subject.
  3. Triple Inconsistency Discovery: Using Open Information Extraction (OLLIE), the system pulls (Subject; Relationship; Object) triples and compares them against a custom Legal Semantic Lexicon of over 17,000 words to find contradictions.

Verb Semantics and Adverbial Modifiers

The success of the verb-centric approach stems from the fact that legal "actions" or "rulings" are typically encapsulated within verbs. By analyzing the relationship between verbs and their modifiers, the system can detect subtle shifts in polarity.

Taxonomy of adverbial modifiers used to detect shifts in verb polarity.

The researchers developed a custom taxonomy of adverbs, categorized by Frequency, Tone, and Manner (see table above). This allows the system to distinguish between descriptions like "elegantly" versus "lazily," which may signal a shift in the court's perspective on a specific action.

Results and Performance

The integration of verb-relationship analysis with the SRI led to a marked improvement in identifying Shift-in-View relationships. While the baseline SRI model failed to identify any such pairs (achieving a precision of 0), the combined system reached a precision of 0.688.

Key findings include:

  • Verb Analysis Dominance: On its own, verb analysis achieved a precision of 0.609, far outperforming sentiment polarity (0.382) and triple inconsistency (0.273).
  • Domain-Specific Sentiment: General-purpose sentiment models are inadequate for legal text, which remains formal and neutral even when delivering contradictory rulings.

Limitations and Future Outlook

Despite high precision, the system faces challenges with recall (0.423), meaning many perspective shifts are still missed due to the complexity of legal phrasing. Furthermore, triple extraction via OLLIE often fails on the long, convoluted sentence structures typical of judicial opinions. Future work will focus on refining legal-specific sentiment annotators and improving semantic similarity measures to better navigate the intricacies of legal jargon.

Summary updated

Abstract

Arguments, counter-arguments, facts, and evidence obtained via documents related to previous court cases are of essential need for legal professionals. Therefore, the process of automatic information extraction from documents containing legal opinions related to court cases can be considered to be of significant importance. This study is focused on the identification of sentences in legal opinion texts which convey different perspectives on a certain topic or entity. We combined several approaches based on semantic analysis, open information extraction, and sentiment analysis to achieve our objective. Then, our methodology was evaluated with the help of human judges. The outcomes of the evaluation demonstrate that our system is successful in detecting situations where two sentences deliver different opinions on the same topic or entity. The proposed methodology can be used to facilitate other information extraction tasks related to the legal domain. One such task is the automated detection of counter arguments for a given argument. Another is the identification of opponent parties in a court case.
BibTeX
@misc{ratnayaka2019shift,
  author = {Gathika Ratnayaka and Thejan Rupasinghe and Nisansa de Silva and Viraj Salaka Gamage and Menuka Warushavithana and Amal Shehan Perera},
  title = {{Shift-of-Perspective Identification Within Legal Cases}},
  year = {2019},
  eprint = {1906.02430},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1906.02430},
}
Figure