Document Type
Poster
Publication Date
Fall 2025
Abstract
We examine how three newspapers in different languages (New York Times in English, Sözcü in Turkish, and NurKZ in Russian) frame the same Syria Trump meeting using aspect based sentiment analysis. Each article was segmented into sentences (NYT 67, NurKZ 40, Sözcü 26), with Turkish and Russian sentences translated into English while preserving the original text for reference. We applied an LLM based ABSA labeler constrained to a fixed set of predefined aspects, allowing us to compare sentiment at the level of specific themes rather than overall tone. Our analysis focuses on four high contrast aspects: sanctions and the Caesar Act, protocol and secrecy, past bounty and terrorism related framing, and stated United States goals or conditions. We measure both how frequently each outlet discusses these aspects and the average sentiment associated with them. The results reveal substantial cross outlet divergence, particularly in sanctions framing, where sentiment ranges from positive in the New York Times to negative in NurKZ, with Sözcü falling in between. We also observe notable differences in emphasis on protocol and secrecy, which receives substantially more attention in Sözcü. To assess reliability, we conducted targeted human annotation on a subset of high contrast sentences, which broadly aligned with the model outputs and supports the validity of the observed patterns.
Recommended Citation
Kocaman, Murathan and Sidorko, Kirill, "A Political Meeting Lost in Translation: Cross-Lingual Aspect-Based Sentiment Analysis of How English, Turkish, and Russian Media Frame the Same Syria-Trump Meeting" (2025). IPHS 391: Interdisciplinary AI Frontiers. Paper 8.
https://digital.kenyon.edu/dh_iphs_391/8
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.