Dataviz: the Bridge between NLP and History ?

Dataviz Workshop

What a couple of days !
It has been a real pleasure to welcome and get to know Benjamin Bach from Edinburgh University. A two-day workshop with participants from quite different disciplines (from history to neuro-linguistics) and different levels of “computer literacy”. For sure it was very challenging invitation for Benjamin who accepted it and did a great job at taking us on this dataviz journey.

From the early (yet surprisingly relevant) theories of Jacques Bertin to very pratical exercices of questioning, sketching and presenting our own datasets, with a hands-on session on the Vistorian and other tools, I now have a lot to digest !

As a realistic dataset, I chose to use the Named Entities extracted from the Boorman Dictionary (those from my previous post) and focused on locations and time. This took me away from NLP and closer to the core objectives of my dear historian colleagues. I left linguistics structures and other machine learning abstractions aside for a moment and tried to figure out how to make things visible in a way which would be useful for the project.

No need for suspense! I did not discover the ultimate and perfect visualization fashion and tool for all our research questions. Nevertheless, Benjamin left us many pieces of advice, good practices, take-home messages and tools to try. We are also looking forward to have more collaboration with him to co-design new ways to explore and present our data.

Concerning the tools, let me try to sum up my temporary conclusions. Firstly we have to find a balance between looking for off-the-shelf tools which can’t fit all our needs, and jumping headlong into writing our own tool from scratch. On the one hand, Padagraph and the Vistorian are great tools, but they may not fit all our need (in other words, we should avoid the “when all you have is a hammer…” mistake). On the other hand, even if I could theoretically implement everything I can sketch, using libs like D3js, there may be a better usage of my time, especially at a stage when we are just exploring a variety of ideas. For that purpose, I’m now looking at higher level libraries like Plot.ly and Vegas. Both happen to be usable from Scala Notebooks to easily produce online editable visualizations for me and my colleagues. I also went back into R-Studio and I’m discovering Shiny, which may be a good option go beyond sketches when brainstorming with colleagues about what we want to visualize. Once we get something satisfying, it will still be time to turn to D3js to implement more convenient interactions for a broader audience.


Pierre Magistry

Pierre Magistry is a postdoctoral researcher in Natural Language Processing with a special interest in Sinitic and low resource languages. He completed his PhD in the ALPAGE team (Paris Diderot – INRIA) in 2013 on Chinese Word Segmentation at the interface of linguistic theories and unsupervised machine learning. His experience benefited from multiple stays in Taiwan (2008-2010, 2014, 2016) under research fundings from TIGP (Academia Sinica), Erasmus Mundus Multi and a Taiwan Fellowship. His work on Taiwanese language led him to release the first input method on mobile devices for this language. In 2017-2018, he was a postdoc at LIMSI (CNRS), working on NLP for low resources situation and developing new semi-supervised learning approaches

You may also like...

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.