Diverse Word Choices, Same Reference: Annotating Lexically-Rich Cross-Document Coreference

Anastasia Zhukova; Felix Hamborg; Karsten Donnay; Norman Meuschke; Bela Gipp

cs.CLFeb 19, 2026

Diverse Word Choices, Same Reference: Annotating Lexically-Rich Cross-Document Coreference

Authors: Anastasia Zhukova, Felix Hamborg, Karsten Donnay, Norman Meuschke, Bela Gipp

Organizations: University of Göttingen Göttingen, Germany · Humboldt University Berlin, Germany · University of Zurich Zurich, Switzerland

Abstract

Cross-document coreference resolution (CDCR) identifies and links mentions of the same entities and events across related documents, enabling content analysis that aggregates information at the level of discourse participants. However, existing datasets primarily focus on event resolution and employ a narrow definition of coreference, which limits their effectiveness in analyzing diverse and polarized news coverage where wording varies widely. This paper proposes a revised CDCR annotation scheme of the NewsWCL50 dataset, treating coreference chains as discourse elements (DEs) and conceptual units of analysis. The approach accommodates both identity and near-identity relations, e.g., by linking "the caravan" - "asylum seekers" - "those contemplating illegal entry", allowing models to capture lexical diversity and framing variation in media discourse, while maintaining the fine-grained annotation of DEs. We reannotate the NewsWCL50 and a subset of ECB+ using a unified codebook and evaluate the new datasets through lexical diversity metrics and a same-head-lemma baseline. The results show that the reannotated datasets align closely, falling between the original ECB+ and NewsWCL50, thereby supporting balanced and discourse-aware CDCR research in the news domain.

Figures & tables

		Annotations
COUNTRY: USA	NewsWCL50	United States; U.S.; White House; CIA director; Matthew Pottinger; Larry Kudlow; chief White House correspondent; American officials; Washington; personal lawyer; C.I.A.; senior director for Asia at the National Security Council; C.I.A. director; U.S. Senators; CIA director and secretary of State-designate; Senate Foreign Relations Committee; director of the East Asia nonproliferation program at the Middlebury Institute; southern White House Director of Trump’s National Economic Council; U.S. officials; F.B.I. director; United Nations Command; His administration; Robert Mueller; Jeffrey Lewis; senators; Kudlow; senior vice president at the Center for Strategic and International Studies; Center for Strategic & International Studies; Special Counsel; Central Intelligence Agency Director; U.S. diplomats; American troops; U.S. secretary of state; United States government; U.S.-led forces; Mr. Trump’s chief economic adviser; CIA chief; National Security Council; Americans; America; Michael D. Cohen; Mr. Kudlow
	NewsWCL50 r	USA : United States; U.S.; White House; his administration; Washington; the U.S.-led forces in the conflict; United Nations Command; the White House press office; the Senate; the U.S.; the Senate Foreign Relations Committee; the Washington-based Center for Strategic & International Studies; the State Department; C.I.A.; signers to the armistice; the government; United States government; Senate Foreign Relations Committee; America; the two countries; the U.N. Command
		USA\Kudlow : Mr. Kudlow; Larry Kudlow, Mr. Trump’s chief economic adviser; Larry Kudlow, director of Trump’s National Economic Council
		USA\Pottinger : Matthew Pottinger, the senior director for Asia at the National Security Council; Matthew Pottinger, senior director for Asian affairs for Trump’s National Security Council
		USA\Tillerson : Tillerson; Rex Tillerson
GROUP: Suffered people	ECB+	t37_victims_of_quake : 24 people; Five; five people; one person; three people

Table I: The annotation example demonstrating the difference between the original annotation schemes of NewsWCL50 and ECB+ to the proposed lexically diverse yet fine-grained annotation scheme.

Dataset	Coreference chains / Discourse elements (DEs)					Mentions				Lexical diversity
Dataset	all	entity	events	singletons	avg. size	all	entity	event	avg. per doc	UL	PD	MTLD
NewsWCL50	134	96	38	4	38.2	5115	3886	1229	102.3	10.71	8.99	14.54
NewsWCL50 r	433	374	59	102	15.1	6531	4758	1773	130.6	6.05	9.06	15.58
ECB+*	171	112	59	59	2.4	407	246	161	16.3	2.19	1.99	4.55
ECB+METAm*	168	109	59	57	2.4	399	240	159	16.0	2.92	3.25	6.82
ECB+ r *	97	84	13	25	14.7	1427	958	469	57.1	5.88	9.04	20.65

Table II: General statistics and lexical diversity metrics for the original and reannotated versions of NewsWCL50 and ECB+. An asterisk (*) denotes that a subset of five subtopics was used.

Figure 1: Distribution of lexical diversity measured by PD and MTLD. Both NewsWCL50 r and ECB+ r exhibit comparable distributions, demonstrating that the proposed annotation scheme achieves a balanced level of lexical diversity in the coreference chains or DEs across news articles, regardless of domain differences.

Dataset	MUC	B3	CEAF e	CoNLL
NewsWCL50	82.49	49.70	11.96	48.05
NewsWCL50 r	79.59	51.82	30.82	54.08
ECB+*	68.35	73.27	68.10	69.91
ECB+METAm*	49.06	67.22	57.85	58.04
ECB+ r *	82.32	47.98	28.46	52.92

Table III: The performance of the same-head-lemma baseline.

Coreference Resolution

Coreference resolution links expressions that refer to the same entities or events across sentences, documents, languages, and modalities.

See all 25 papers

Piecing Together Cross-Document Coreference Resolution Datasets: Systematic Dataset Analysis and Unification
Feb 28, 2026Anastasia Zhukova, Terry Ruas, Jan Philip Wahle +1Coreference Resolution Cross-Dataset Generalization
University of Göttingen Göttingen, Germany
Introducing corpora Hlava Cor and Hlava AD: Human Label Variation in Coreference and Discourse Relations
Jun 24, 2026Anna Nedoluzhko, Šárka Zikánová, Jiří Mírovský +2Inter-Rater Reliability Coreference Resolution
Charles University, Faculty of Mathematics and Physics, Institute of Formal and Applied Linguistics, Prague, Czech Republic
CorPipe at CRAC 2026: Empty Nodes and Cross-Lingual Transfer in Multilingual Coreference Resolution
May 28, 2026Milan StrakaCoreference Resolution Cross-Lingual Transfer
Charles University, Faculty of Mathematics and Physics Institute of Formal and Applied Linguistics Malostranské nám. 25, Prague, Czech Republic

Feb 28, 2026J/K move · Enter open · S save

Piecing Together Cross-Document Coreference Resolution Datasets: Systematic Dataset Analysis and Unification

Anastasia Zhukova, Terry Ruas, Jan Philip Wahle, Bela Gipp

University of Göttingen Göttingen, Germany

Coreference Resolution Cross-Dataset Generalization

Open paper AskPDF arXiv

Research in CDCR remains fragmented due to heterogeneous dataset formats, varying annotation standards, and the predominance of the CDCR definition as the event coreference resolution (ECR). To address these challenges, we introduce uCDCR, a unified dataset that consolidates diverse publicly available English CDCR corpora across various domains into a consistent format, which we analyze with standardized metrics and evaluation protocols. uCDCR incorporates both entity and event coreference, corrects known inconsistencies, and enriches datasets with missing attributes to facilitate reproducible research. We establish a cohesive framework for fair, interpretable, and cross-dataset analysis in CDCR and compare the datasets on their lexical properties, e.g., lexical composition of the annotated mentions, lexical diversity and ambiguity metrics, discuss the annotation rules and principles that lead to high lexical diversity, and examine how these metrics influence performance on the same-head-lemma baseline. Our dataset analysis shows that ECB+, the state-of-the-art benchmark for CDCR, has one of the lowest lexical diversities, and its CDCR complexity, measured by the same-head-lemma baseline, lies in the middle among all uCDCR datasets. Moreover, comparing document and mention distributions between ECB+ and uCDCR shows that using all uCDCR datasets for model training and evaluation will improve the generalizability of CDCR models. Finally, the almost identical performance on the same-head-lemma baseline, separately applied to events and entities, shows that resolving both types is a complex task and should not be steered toward ECR alone. The uCDCR dataset is available at https://huggingface.co/datasets/AnZhu/uCDCR, and the code for parsing, analyzing, and scoring the dataset is available at https://github.com/anastasia-zhukova/uCDCR.

Abstract

Coreference Resolution

Explore similar work

Piecing Together Cross-Document Coreference Resolution Datasets: Systematic Dataset Analysis and Unification

Introducing corpora Hlava Cor and Hlava AD: Human Label Variation in Coreference and Discourse Relations

CorPipe at CRAC 2026: Empty Nodes and Cross-Lingual Transfer in Multilingual Coreference Resolution