During this project, I worked with a cross-institutional team building one of the first large-scale datasets to explore how language shapes the outcomes of real-world negotiations. The dataset consisted of controlled behavioral lab experiments where participants negotiated house prices, either through simple numerical offers or open, audio-based conversations. I analyzed over 6,000 minutes of recorded negotiations using Python, where I engineered sentiment analysis pipelines, built word embeddings, and annotated speech patterns to help capture the linguistic behaviors driving these high-stakes conversations.
Through my analysis, we uncovered that while final agreed prices stayed consistent across both communication styles, conversations with natural language led to faster agreements, higher success rates, and less variability in outcomes. The sentiment analysis I ran revealed how effective buyers leveraged personal concerns and emotional framing to push for better prices, while sellers who used targeted questions were more likely to steer conversations to their advantage. The speech pattern annotations showed how subtle linguistic behaviors, such as comparisons and push strategies, could predict negotiation success early in the interaction.