What You Say (WYS): Automated Two-Level Pipeline for Interpretive Analysis of Digital Conversation
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Abstract The growing availability of exported chat data has created new opportunities for behavioral researchers to examine communication patterns in naturalistic text. However, currently visible tools tend to split into two unsatisfactory extremes: lightweight consumer dashboards that provide only descriptive summaries, and research-oriented pipelines that require technical setup and do not generate accessible user-facing reports. We present What-You-Say (WYS), an automated two-level pipeline for applied conversation analysis that separates deterministic feature extraction from bounded large language model (LLM) synthesis. In Level A, the system parses exported chats and computes reproducible lexical and structural indicators from pre-specified dictionaries and rule-based procedures. In Level B, these extracted signals are translated into a constrained interpretive report designed for readability while remaining anchored to the Level A outputs. The methodological contribution of WYS is this explicit separation of measurement and interpretation: identical text produces identical Level A outputs, whereas Level B provides bounded synthesis rather than acting as an independent measurement engine. We describe the architecture, situate the tool relative to existing chat-analysis systems, and illustrate its use through worked examples from the live application. The paper is intended as a tool and instrumentation article rather than a formal external validation study. WYS is presented as a reproducible framework for accessible chat analysis that may support future research on personal, dyadic, and group communication. Keywords: conversation analysis; computational text analysis; natural language processing; large language models; reproducibility; research software.
Fields of Study & Subjects
Computational and Text Analysis Methods
Language
Discourse
Communication Strategies
Sentiment Analysis and Opinion Mining
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 26 March 2026
- Language
- EN
- Document type
- Preprint