Conclusions, project implications, next steps

My main conclusions about the project’s guiding question are noted in the discussion section. Here, I explore what the project taught me about using topic modelling, about speeches from the throne as a corpus, and about the combination of the two for the study of government priorities. I also expand on some potential next steps.

What can we learn about…

…topic modelling?

Topic modelling is a useful tool to quickly get a sense of a corpus. Some prior knowledge is certainly beneficial to help determine the number of topics and to get a sense of what the topics mean. But for those with little to no knowledge of an historical period, topic modelling can quickly reveal interesting points on which to dive in, such as the 1973 oil crisis, by identifying major points of discussion at certain times. However, one must be careful not to draw unfounded conclusions: the various mentions of the Arctic between 2006 and 2015, for example, likely had less to do with a particular historical event than with the political interests of the governing party at the time.

Topics like “declining parliamentary formality” reveal the great potential of topic modelling as a tool to suggest research opportunities. Admittedly, it is not a particularly relevant topic to this project’s overall question. But it suggested an entirely new path of research—thanks to topic modelling, I was better able to think through the corpus and understand its possibilities for research.

…speeches from the throne?

They do offer a decent corpus for analyzing priorities. We were able to observe, for example, the rise in interest in social policy as a broad issue area. That said, they are not an exhaustive listing of a government’s priorities, and many more nuanced or niche issues may not have been picked up by the topic modeller.

That said, the topic modeller would have been more effective with a larger corpus. 158,000 words and 51 documents is relatively small for topic modelling. It may have been more fruitful to use a corpus like yearly budget speeches, and to expand the years under analysis. However, for the purposes of this small project, speeches from the throne worked well enough.

…government priorities?

Ultimately, the topics generated did not surface too many “issue” topics. Given the constraints of topic modelling, noted above, this is not necessarily a surprise.

The approach taken by Montpetit and Foucault (2012), tagging portions of sentences with specific policy issues, would be more effective at exhaustively identifying the priorities discussed by governments in speeches from the throne. However, such an approach may not have yielded the alternative conclusions suggested by topic modelling, and it would have taken far more time.

Nonetheless, we were able to observe various trends, as noted in the discussion section. While this corpus and tool could not provide an extremely detailed portrait of such trends, it was certainly able to point to some shifts.

Next steps

This work suggested a variety of paths for future research: