Our own Prof Will Lamb is working with Dr David Howcroft (lead investigator) and Dr Dimitra Gkatzia from Edinburgh Napier university to build the first tools for Gàidhlig chatbots. This is starting with the creation of a new dataset to train AI models.
Our current experiments (which you can participate in if you speak Gàidhlig!) are focused on building our dataset: we need examples of humans asking and answering questions about museum exhibits in a chat conversation. Participants are paired up and given a set of exhibits from the National Museum of Scotland to discuss, briefly summarising their discussions as well.
The next step? Well, after a bit of data cleanup and anonymisation, it’s time to see how well neural network models for natural language generation work for this amount of data. One of the interesting challenges for this project is trying to see how far you can get in building a chatbot with as little data as possible. The lessons we learn in this work will inform future work, not just in Scottish Gaelic, but in Natural Language Generation more generally!
Why build chatbots for Scottish Gaelic?
We believe the world is a better place when everyone can learn in their preferred language. Scottish Gaelic has fewer language technologies available than languages like English or Mandarin, and we’d like for our research in natural language generation to help in some small way to address this gap.
Why focus on ‘Exhibits’?
Museums are a primary tool for learning outside of schools, libraries, and documentaries, and are increasingly leveraging mobile applications and chatbots to enhance visitor experiences. However, these chatbots are generally available for only a few languages, due to a lack of linguistic and technical resources for minority languages like Scottish Gaelic.
How can I contribute?
If you speak Scottish Gaelic and live in Scotland, you can take our short comprehension quiz (5-10 minutes) and sign up to participate in the study! The full study (after the quiz) takes up to two hours to complete, and participants will receive up to £30 in compensation for their contribution. Additionally, you’ll have the opportunity to be named as contributing to this important Gaelic resource if you so desire! More details here: https://nlg.napier.ac.uk
If you don’t speak Scottish Gaelic or live outside of Scotland, you can share this blogpost with all the Gaelic speakers you know! Encourage them to participate or to spread the word to their friends. All in all, we hope to recruit about 100 people to participate in our study, and we have a ways to go before we reach this goal. If you don’t know what to say to your contacts, how about:
Researchers in Edinburgh are trying to build the first chatbots for Scottish Gaelic and they’re recruiting participants for an experiment paying up to £30! Find out more at: https://blogs.ed.ac.uk/garg/2022/08/23/scottish-gaelic-chatbots-for-museum-exhibits/ or sign up at https://nlg.napier.ac.uk
Who all is involved in this research?
This work is supported by a small grant from Creative Informatics. The lead investigator is Dave Howcroft. William Lamb and Dimitra Gkatzia are co-investigators. Anna Groundwater from the National Museum of Scotland provided information about their exhibits, and Hector Michael Fried & Rory Gianni (InChat Design) support the effort as well. We are also grateful to our student intern for their assistance.