Summary:

This internal event is part of our AI and the Arts series. It will highlight and map out expertise and strengths at the intersection of AI / data science and languages / linguistics at the University of Manchester. This highly interdisciplinary area opens up a number of exciting possibilities for research, teaching, knowledge exchange, and business engagement.

Researchers from across the University of Manchester’s Digital Futures and Creative Manchester networks will spotlight their research in these areas.

Speakers

• Dr Colin Bannard - Senior Lecturer in Linguistics, University of Manchester
• Dr Ricardo Bermúdez-Otero - Senior Lecturer in Linguistics and English Language
• Prof Tobias Galla - Professor of Theoretical Physics
• Dr Andrea Nini - Lecturer in English Language, University of Manchester

Talks

Using computational linguistics to detect markers of Parkinson's disease in typing data Dr Colin Bannard - Senior Lecturer in Linguistics, University of Manchester

I will describe work on using computational linguistics in order to identify a marker of Parkinson's disease in keystroke recordings from typing. An often reported early sign of the Parkinson's disease is difficulty in performing habitual behaviours. In this work we analyse samples of a highly automatised behaviour - language use, or specifically typing - and show a reduction in two key behavioural signatures of habit can be used to distinguish people with a recent diagnosis of Parkinson's disease from controls.

A voter-model approach to the distribution and dynamics of typological features of language

Dr Ricardo Bermúdez-Otero - Senior Lecturer in Linguistics and English Language and Prof Tobias Galla - Professor of Theoretical Physics

The world’s languages can be classified according to a wide range of features: e.g. whether they place the verb before or after the object (VO vs OV word-order), whether or not they have a definiteness marker akin to English _the_, etc. Features are passed from earlier to later periods in the history of each language, and also across neighbouring languages. As this happens, different features change at different rates: e.g. the OV word-order is comparatively stable, whereas definiteness markers come and go more frequently. In previous work [1], we have shown that a feature’s relative propensity to change can be estimated just by looking at whether the languages that have it are clumped or scattered in geographic space. This is made possible by a model of language change that uses tools first developed in statistical physics and probability theory.

We are currently developing this programme of research by means of an APEX award from the Royal Society and the British Academy [2]. At present our attention is focused on how linguistic features interact with one another: we know, for example, that the chances of word-order flipping from OV to VO go up if the language has prepositions instead of postpositions. In future work, we will also consider how physical barriers affect the distribution of linguistic features in geographic space, as such barriers constrain the processes of migration and diversification through which languages split into multiple descendants. Finally, to explain a feature's propensity to change across the world's languages, we plan to draw on physical, cognitive and social factors acting on individual speakers.

This work may provide a template for using geospatial data to understand other forms of cultural evolution.

[1] Kauhanen, Henri, Deepthi Gopal, Tobias Galla & Ricardo Bermúdez-Otero. 2021. Geospatial distributions reflect temperatures of linguistic features. _Science Advances_ 7 (1), eabe6540.

[2] https://royalsociety.org/news/2021/09/apex-2021-announced/

Computational authorship identification

Dr Andrea Nini - Lecturer in English Language, University of Manchester

In this talk I will give an overview of modern computational approaches for the identification of anonymous writers, especially for forensic and investigative purposes. I will also cover how the application of these techniques is providing new solutions for investigators working on cybercrime on the Dark Web.

Find out more:

You can watch all the presentations from our first AI and the Arts event here.

This is a cross-platform event between the Digital Futures Creative & Heritage and Data Science & AI themes and the Creative Manchester platform.

Sign up to the Digital Futures newsletter and the Creative Manchester newsletter to find out more about activity across both platforms.