Narrative Consolidation: Formulating a New Task for Unifying Multi-Perspective Accounts

Processing overlapping narrative documents, such as legal testimonies or historical accounts, often aims not for compression but for a unified, coherent, and chronologically sound text. Standard Multi-Document Summarization (MDS), with its focus on conciseness, fails to preserve narrative flow. This paper formally defines this challenge as a new NLP task, Narrative Consolidation, whose central objectives are chronological integrity, completeness, and the fusion of complementary details, and establishes the resources needed to study it: a formal task definition, an evaluation paradigm that includes a selection-level metric, the Gospel Consolidation Language Resource --- a benchmark built from the four Biblical Gospels, with 169 canonical events, cross-document alignments, and a manually created reference consolidation --- and a suite of reference systems, ranging from a timeline-agnostic centrality method to timeline-aware heuristics and the Temporal Alignment Event Graph (TAEG), a multi-relational graph that explicitly models chronology and event alignment. Benchmarking these systems yields three findings that characterize the task. First, the explicit temporal backbone is the dominant factor: every system granted the canonical timeline raises ROUGE-L F1 from 0.206 to at least 0.81, whereas content-selection sophistication accounts for a far smaller margin. Second, on a fusion-style reference, a simple length heuristic --- selecting the longest available account of each event --- is remarkably strong (0.947 ROUGE-L F1; 0.648 selection accuracy), outperforming the graph-based selection (0.846; 0.489), which is nevertheless clearly above the random floor. Third, ablations show that the discriminative signal resides in the temporal edges, while intra-cluster lexical similarity is uninformative --- a negative result that constrains the design of future models. Together, these results establish Narrative Consolidation as a task distinct from summarization, provide the first reference points for it, and pose an explicit open challenge: to surpass that heuristic with a principled selection mechanism, and to move beyond version selection towards the fusion of complementary details.