Do you like colourful bubble charts? Then you’re in luck, because we are introducing a new feature to our site that involves exactly that! The feature will be rolled out in phases, and new aspects will be added to it in phases, but you can already use it now.
Introducing: our keyword function!
Our keyword function displays in a bubble chart which words in one part of Hansard compared to another part of Hansard. This is modelled on the keyword function often used in corpus linguistics, where keywords are often used in corpus linguistics to answer questions like: “What is the body of text mainly about?”, “What are the major themes or concerns of the text?” and “What makes a text distinct from another text?”. Keywords computed with this function are those words which occur relatively more frequently in one corpus compared to another corpus. To do this, we compare the corpus we are interested in knowing keywords for (the ‘target corpus’) with a ‘comparison’ corpus. The algorithm behind the comparison then calculates which words are key by carrying out a statistical test called ‘Log-likelihood’. The higher the log-likelihood score, the more confident we can be that the frequency difference between the two corpora is not just by chance. The keyword function on our site only shows the keywords of which we are 99.99% certain this difference is not by chance. The words you will find in the bubble chart are thus those that we are very certain that the difference in word frequency between the target and comparison corpora is is statistically significant.
You can either self-define the two sections of Hansard you want to compare or you can select pre-programmed time periods (such as decades and various wars). One of the ways in which we are planning to develop the keyword function is to add to the various pre-programmed sections of Hansard.
From your bubble chart you can select up to four items to search for them within your chosen target corpus. The four terms will be shown in concordance lines just as everywhere else on the site.
If you have any questions about this feature, please get in touch! We would love to explain more!
