Joint Seminar l Spinning the discourse roulette: addressing poverty of data with Monte Carlo simulations

An intractable problem for analysts of professional communication and discourse is not having enough data due to confidentiality or other logistical hurdles. This poses additional challenges if findings are to be seen as useful and applicable. In this talk, I share my recent exploratory work with Monte Carlo simulations (MCS), a class of computational methods used to address similar problems in fields like finance and gaming (Kroese et al., 2014), but almost never on language data. The basic logic of MCS is to i) simulate repeated random samples of a phenomenon - language and discourse in this case - using limited information about its probability distribution, and ii) use the simulations to estimate how various discourse scenarios would have panned out anyway by the laws of chance. I illustrate this with psychotherapy, a professional context where linguistic analysis is crucial but data is hard to obtain. I show how i) therapy talk is first quantified under socio-psychological and linguistic categories like analytic thinking, authenticity, clout, emotional tone, and pronouns (Pennebaker et al., 2015), ii) MCS models are then built and validated with ’training’ and ’test datasets’, iii) accurate models can inform follow-up qualitative/quantitative analysis, as well as therapists’ understanding of the linguistic behavior of clients. The whole process is implemented with Python, an open-source programming language widely used in industry and research today.

References

Kroese, D. P., Brereton, T., Taimre, T., and Botev, Z. I. (2014). Why the Monte Carlo method is so important today. Wiley Interdisciplinary Reviews: Computational Statistics, 6(6):386–392.

Pennebaker, J.W., Boyd, R. L., Jordan, K., and Blackburn, K. (2015). The development and psychometric properties of LIWC2015. University of Texas at Austin, Austin, TX.