“Popular tribunes” and their agendas: topic modelling Slovak presidents’ speeches 1993–2020

ABSTRACT

Since its birth as an independent republic in 1993, Slovakia has been served by five different presidents. Due to limited competences, the presidents' have often relied on political speech as their principal tool to influence political developments. However, text as a source of data has been largely neglected in existing scholarship on Central European presidents. In this exploratory study, I classify the content of presidential speeches using a topic model and analyse topical patterns over time and across different presidents. I find that topical variation can provide useful insights into relevant issues such as agenda shifts or intra-executive conflict.

KEYWORDS:

• Slovakia
• presidents
• political speech
• quantitative text analysis
• non-negative matrix factorisation

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Sedelius and Ekman (2010, 512) posit that in transitioning democracies, the personal authority of the office bearer is greater due to incomplete institutionalization.

2 These performances were still read outs of written speeches, delivered with few instances of misspeaking, so discrepancies between the original speech and the transcript should be minimal. Hesitations and distractions were not recorded in the transcript.

3 Because translation of some texts is liable to introduce bias into the corpus, it is not generally recommended. However, owing to the small proportion of translated texts (0.02%), I deemed the inclusion of these texts an acceptable trade-off between more data and less reliability. Removing them instead does not significantly affect the results.

4 An interesting way of extending the model would be to weight observations by importance or audience size of each speech. However, it is difficult to find suitable data that could operationalize this variable.

5 Sentences are a useful unit for this type of pre-processing, because we are less likely to introduce stray words about different topics than if the speeches were partitioned purely by number of words disregarding sentences completely.

6 Similarly, the results are not dramatically different when using LDA instead of NMF to estimate the word-topic coefficients and document-topic proportions.

7 English terms appear where a multi-word expression was collapsed into a single term (for example “westernbalkans”).

8 President Gašparovič‘s address at the national union of employers, Bratislava, 26 April 2013.

9 Alternative smoothing algorithms, such as locally estimated scatterplot smoothing, yield similar trend lines.

10 Estimated topic prevalence $\left(\hat{\theta }\right)$ for topic “government”: ${\hat{\theta }}_{CAP}=0.055,{\hat{\theta }}_{KIS}=0.040,{\hat{\theta }}_{GAS}=0.021,{\hat{\theta }}_{SCH}=0.058,{\hat{\theta }}_{KOV}=\mathrm{0.072.}$ While topic prevalence and top word counts are not interchangeable measures, the latter might be more tangible for the reader. The proportion p of the word “government” in the vocabulary (frequency of all words used) of presidents is as follows: $p_{CAP}=0.0043,p_{KIS}=0.0041,p_{GAS}=0.0027,p_{SCH}=0.0058,p_{KOV}=\mathrm{0.0060.}$

11 ${\hat{\theta }}_{people}:{\hat{\theta }}_{CAP}=0.054,{\hat{\theta }}_{KIS}=0.069,{\hat{\theta }}_{GAS}=0.023,{\hat{\theta }}_{SCH}=0.013,{\hat{\theta }}_{KOV}=\mathrm{0.019.}$ Proportion p of the word “people” in the vocabulary: $p_{CAP}=0.006,p_{KIS}=0.011,p_{GAS}=0.003,p_{SCH}=0.002,p_{KOV}=\mathrm{0.003.}$

12 ${\hat{\theta }}_{judiciary}:{\hat{\theta }}_{CAP}=0.052,{\hat{\theta }}_{KIS}=0.043,{\hat{\theta }}_{GAS}=0.005,{\hat{\theta }}_{SCH}=0.001,{\hat{\theta }}_{KOV}=\mathrm{0.001.}$ Proportion p of the words “court” and “judicial” in the vocabulary: $p_{CAP}=0.009,p_{KIS}=0.007,p_{GAS}=0.002,p_{SCH}=0.002,p_{KOV}=\mathrm{0.002.}$

Additional information

Notes on contributors

Michal Ovádek

Michal Ovádek is a PhD candidate at the Centre for Legal Theory and Empirical Jurisprudence at KU Leuven. His research focuses on the interplay between law and politics in various institutional contexts, in particular the European Union. His work appeared in, among others, European Union Politics, European Journal of International Law, Journal of European Integration and European Journal of Law and Economics.