Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study investigates the classification of fine-grained inconsistencies in financial disclosure texts, focusing on how different types of inconsistencies—numerical, temporal, referential, factual, and normative—require distinct evidence for accurate identification. By evaluating various models, including frozen embedding classifiers and fine-tuned encoders, on a synthetic benchmark of 5,940 instances, the authors find that a fine-tuned 300M encoder achieves the highest accuracy of 65.3% when supplied with gold evidence spans. The results highlight the critical role of localization quality in improving classification performance, particularly for referential inconsistencies, and suggest that advancements in evidence extraction and reasoning are essential for future progress.
Fine-tuned encoders outperform larger models in classifying financial disclosure inconsistencies, revealing the importance of evidence quality over sheer model size.
Financial disclosures contain numerical claims, temporal statements, entity references, policy commitments, and risk descriptions that may conflict in qualitatively different ways. Detecting a conflict is only the first step: review workflows may also need to determine its type, since numerical, temporal, referential, factual, and normative inconsistencies require different evidence and downstream checks. We study this problem as fine-grained inconsistency classification. Using a fixed 5,940-instance snapshot of SBID-FD, a synthetic financial-disclosure benchmark with 11 inconsistency labels and paired reference evidence spans, we compare frozen embedding classifiers, fine-tuned encoders, evidence-augmented classifiers, prompted large language models, and LoRA-adapted generative models under a shared evaluation protocol. A fine-tuned 300M encoder reaches 61.9% accuracy, compared with 61.5% for a LoRA-adapted Qwen3.5-9B model and 61.3% for GPT-5.4. Because these systems differ in architecture, supervision, training objective, and input format, we interpret this as a practical efficiency result for compact supervised encoders rather than a controlled conclusion about model scale. Supplying gold evidence spans improves the fine-tuned encoder to 65.3%, whereas automatically predicted spans recover a meaningful but incomplete share of that gain, indicating that localization quality remains a bottleneck. Class-level analyses show that Referential inconsistencies are especially sensitive to localization quality, while Factual and Logical inconsistencies remain difficult even when the relevant evidence is provided. Together, the oracle, distractor, and per-class analyses separate localization errors from residual type-discrimination errors, indicating that progress requires both stronger evidence extraction and better reasoning over closely related inconsistency categories.