Supporting data for: LLM-Assisted Keymorph Analysis of Grammatical Case in RT's Israeli–Palestinian Conflict Coverage

Dataset description: The dataset for this study supports a Keymorph Analysis of grammatical cases in Russian-language news headlines concerning the the 2023-2025 Israeli-Palestinian conflict, collected from RT's official news website. The dataset comprises four main components: Raw Headlines and Filtered Corpus: This component includes the initial collection of Russian-language headlines from RT (2023-10-07 to 2025-01-19) and the subsequently filtered corpus of 8,757 distinct headlines containing specified keywords related to the conflict (e.g., 'Israel', 'Palestine', 'Gaza', 'Hamas'). Reference Corpus: The reference corpus was constructed from the National Media Subcorpus of the Russian National Corpus (RNC). Annotated Corpus of Grammatical Cases: This core component features the grammatical case annotations for 11 identified target keywords across the corpus. The annotations were generated using an LLM (ChatGPT-5 mini API) with a 20% human-reviewed and corrected sample integrated into the final dataset to ensure high quality and accuracy. Derived Analytical Data and Visualizations: This includes statistical summaries of keyword frequencies and grammatical case distributions, standardized Pearson residual values and log-likelihood (LL) ratio values crucial for keymorph identification, and various visualizations such as word frequency charts and residual heatmaps, all derived from the annotated corpus to support the keymorph analysis. Related article abstract: This study applies and extends Keymorph Analysis (KMA) with cognitive linguistic theory to investigate the representation of the Israeli–Palestinian conflict in Russia Today (RT)’s Russian-language headlines. Unlike traditional keyword analysis, which primarily focuses on lexical content, KMA reveals underlying narrative orientations by examining how systematic morphosyntactic choices contribute to the construal of participant roles. Our approach integrates three analytical layers: (1) a Quantitative Layer that identifies statistically significant keymorphs using a novel dual-reference framework (Standardized Residuals for internal distinctiveness and Log-likelihood tests against a broad reference corpus) via LLM-enhanced annotation (98.58% accuracy); (2) a Contextual Analysis Layer that maps these grammatical patterns to their specific lexical and semantic environments through corpus-assisted analysis; and (3) a Cognitive-Semantic Interpretation Layer grounded in the cognitive-semantic networks of the Russian case system. Through this integrated analysis, we identify a core-periphery hierarchy in case usage, revealing three contrastive cognitive schemas: military agents vs. humanitarian space, active entities vs. constrained subjects, and external dominance vs. regional passivity. Ultimately, this study provides a scalable, LLM-enhanced methodology for analyzing morphologically rich languages, advancing our understanding of how grammatical case assignment functions as a systematic mechanism for organizing participant positioning and constructing divergent narrative framings.本研究使用的数据集支持针对2023-2025年巴以冲突相关俄语新闻标题的关键词形态分析（Keymorph Analysis），数据采集自RT官方新闻网站。该数据集包含四个核心组成部分： 1. 原始标题与过滤语料库：该部分包含2023年10月7日至2025年1月19日期间从RT采集的初始俄语新闻标题集合，以及后续过滤得到的8757条包含冲突相关指定关键词（如"以色列"、"巴勒斯坦"、"加沙"、"哈马斯"）的独特标题语料库。 2. 参考语料库：该参考语料库构建自俄罗斯国家语料库（Russian National Corpus, RNC）的国家媒体子语料库。 3. 语法格标注语料库：作为核心组件，该部分为语料库中11个已识别的目标关键词提供语法格标注。标注工作通过大语言模型（Large Language Model, LLM）ChatGPT-5 mini API完成，同时将20%经人工审核修正的样本纳入最终数据集，以保障标注质量与准确性。 4. 衍生分析数据与可视化成果：该部分包含关键词频率与语法格分布的统计摘要、用于关键词形态识别的标准化皮尔逊残差值与对数似然比（Log-likelihood, LL）值，以及各类可视化成果（如词频图表、残差热图），所有内容均从标注语料库中衍生而来，用于支撑关键词形态分析。 相关论文摘要：本研究将关键词形态分析（Keymorph Analysis, KMA）与认知语言学理论相结合，探究RT俄语新闻标题对巴以冲突的叙事呈现。与传统主要聚焦词汇内容的关键词分析不同，关键词形态分析通过考察系统形态句法选择如何影响参与者角色的构建，揭示潜在的叙事倾向。本研究整合了三层分析框架：（1）定量分析层：通过大语言模型增强的标注流程，采用新颖的双参考框架（基于内部显著性的标准化残差分析、针对宽泛参考语料库的对数似然检验）识别具有统计显著性的关键词形态，标注准确率达98.58%；（2）语境分析层：通过语料库辅助分析，将这些语法模式映射至其特定的词汇与语义语境中；（3）认知语义阐释层：基于俄语格系统的认知语义网络开展阐释。通过该整合分析，本研究识别出格使用中的核心-外围层级结构，并揭示了三组对比性认知图式：军事行动主体与人道主义空间、主动实体与受限主语、外部主导与区域被动。最终，本研究提出了一种可扩展的、基于大语言模型增强的分析方法，用于分析形态丰富型语言，增进了我们对语法格指派如何作为系统性机制来组织参与者定位与构建差异化叙事框架的理解。