Speaker: Natalia Levshina, Linguist at the Max Planck Institute for Psycholinguistics in Nijmegen

Natalia is a linguist working at the Max Planck Institute for Psycholinguistics in Nijmegen. Her main research interests are cognitive and functional linguistics, pragmatics, typology, corpora and data science. She obtained her PhD at KU Leuven in 2011 and got her habilitation qualification at Leipzig University in 2019 with a thesis Towards a theory of communicative efficiency. In addition to papers on causatives, differential case marking, politeness, word order variation and other linguistic topics, Natalia is the author of a best-selling statistical manual How to Do Linguistics with R (Levshina 2015).

Title: Recycle, relax, repeat: Advantages of Bayesian inference in comparison with frequentist methods

Abstract: 

Bayesian inference is becoming increasingly popular in linguistic research. In this talk I will compare frequentist (maximum likelihood) and Bayesian approaches to generalized linear mixed-effects regression, which is de facto the standard method for testing linguistic hypotheses about linguistic variation. The main advantages of Bayesian inference include an opportunity to test the research hypothesis directly, instead of trying to reject the null hypothesis. One can also use information from previous research as priors for subsequent models, which helps to overcome the recent crisis of reproducibility. This also enables one to use smaller samples. It helps to solve such problems as overfitting, data separation and convergence issues, which often arise when one fits generalized mixed-effect models with complex structure. These advantages will be illustrated by a multifactorial case study of help + (to-)infinitive in US magazines, as in the example These simple tips will help you (to) survive the Zombie apocalypse.

About The Language Technology and Data Analysis Laboratory (LADAL) Webinars 2021

The Language Technology and Data Analysis Laboratory (LADAL) is school-based support infrastructure for computational humanities research established and maintained by the UQ School of Languages and Cultures. The LADAL is part of the ARDC Australian Text Analytics Platform (ATAP) which represents a nation-wide attempt to foster computational skills in HASS. It collaborates with and shares expertise with several Australian and international centres, institutions, researchers, and experts.

The LADAL consists of a specialist computing lab for language-based computational and experimental work (the Computational and Experimental Workshop) and an online virtual lab (the LADAL website). The LADAL website offers self-guided study materials and hands-on tutorials on topics relating to digital tools, computational methods for data extraction and processing, data visualization, statistical analyses of language data, and provides links to further resources and short descriptions of digital tools relevant for digital HASS research. In addition, the LADAL offers face-to-face consultations and specialized workshops. SLC researchers are encouraged to contact LADAL staff for advice and guidance on matters relating to digital research tools, data visualization, various statistical procedures, and text analytics.  As such, the LADAL offers pathways to new research possibilities in HASS with a focus on computational quantitative text analytics.