Whenever I speak to my dad, I inevitably get drawn into long conversations on Singapore politics. Like many Singaporeans his age, my dad loves to complain about how today’s 4G leaders are not as competent as Lee Kuan Yew (henceforth LKY) and his first-generation cabinet.
One day, I finally bit the bullet. “How exactly are our current leaders different?” I asked my dad. In a trice, he whipped out his phone to play a YouTube video clearly bookmarked in anticipation of this moment. The video begins with LKY’s now-famous, or infamous, depending on who you ask, speech after breaking up a Singapore Airlines Pilots’ Association strike. The pilots were demanding a 30 percent pay hike and Lee answered their demands in person, threatening to ground the entire airline and drag the pilots through the mud. In the speech, Lee delivers a performance for the ages. Jaw clenched, every muscle in his body taut and right index finger jabbing the air menacingly, he snarls: “Whoever governs Singapore must have that iron in him, or give it up! This is not a game of cards. This is YOUR life and mine!”
My dad pauses the video and lectures me, “You see how he speaks? Now you see how this clown talks!” The video resumes and cuts to Lee Hsien Loong at a 2015 General Election (GE) rally repeating his father’s iconic line. But something is amiss. The junior Lee is clearly struggling to channel his father’s gusto. The wrong syllables are emphasised and the delivery stilted; he sounds more like an awkward teenager than an iron-willed statesman. My dad, having made his case with a single cherry-picked comparison, smirks, “Look at them, can only copy, no originality”. The comments under the video echo the same sentiment.
Undoubtedly, I feel a certain sympathy for Lee Hsien Loong in this situation. Living under his father’s shadow was bound to result in embarrassing moments such as this. Yet, the video does raise the interesting question of whether there are indeed distinct rhetorical differences among Singapore’s political leaders. Differences so acute that they end up inadvertently shaping broader voter perceptions of Singapore politics.
The study of political rhetoric has a rich tradition, drawing scholars from political science, linguistics and anthropology. Research in this field has traced the evolution in the speaking styles of politicians across time, polity and language. From the “enigmatic language” of Mao Zedong, consisting of intentionally vague sentences designed to serve as self-mythologising riddles to Franklin D Roosevelt’s famous fireside chats, meticulously constructed to soothe an anxious electorate gripped by the Great Depression while simultaneously criticising his political opponents, rhetorical analysis helps us understand how politicians use language to achieve a variety of ideological objectives.
Indeed, our historical memory of LKY as a visionary statesman is heavily intertwined with our collective recollection of him as a uniquely gifted orator. Research papers have been written about Lee’s charismatic leadership and the World Economic Forum has an entire page dedicated to his “eloquence”. By contrast, little such praise has been showered upon Lee Hsien Loong. In fact the top result on googling “Lee Hsien Loong’s charisma” is a speech in which he states that “[Singapore] still needs charismatic leaders like [Lee Kuan Yew]”.
This difference in scholarship is not without merit. Scholars have pointed out that LKY developed narratives around “Asian Values” and “Confucian Ethics”, allowing him to shield Singapore from external critique by reifying differences between the country’s electoral autocracy and Western liberal democracy. While Lee Hsien Loong and Goh Chok Tong too wielded these narratives in their speeches, their rhetoric rested on the foundations laid by Lee Senior.
Meanwhile, contemporary analyses of Singapore’s political rhetoric has shown that current leaders speak primarily in corporate platitudes designed to attract capital from the global market. Their primary motifs are stability and efficiency, best described as “boring in a capillary fashion”. And while this monotony undoubtedly reflects a prioritisation of substance over style, the former does not preclude the latter. If anything, Lee Hsien Loong’s channeling of his father’s words indicates that our political class still desires rhetorical flourish, especially during election season.
This begets the question of just how much Singapore’s political rhetoric has shifted since LKY. Are our current leaders uniquely uncharismatic compared to Lee? And are they becoming increasingly similar and consequently boring? Is the popularity of Tharman Shanmugaratnam partially explainable due to his superior oratory? Advances in computational analysis can help us answer these questions.
Understanding changes in rhetoric and similarities requires an analysis over time and across speakers. This is done through stylometry, a statistical technique first devised to determine authorship of previously unidentified texts, most famously in the case of the Federalist Papers. Stylometry focuses on the types of words used and their relative frequencies to create word count distributions. It then extends these capabilities by including combinations of words used and their frequencies to construct a multi-dimensional description of a speech. However, stylometry is not a substitute for deep reading. Rather, it is a complementary method that can analyse texts at scale and point researchers in promising directions.
In our case, stylometry is appropriate and useful for three reasons. First, because we already know who the authors of these speeches are, we can compare just how different they are lexically. Second, we are not as concerned with the substance of the speeches as we are with their rhetorical style. Thus, focusing on subject-verb-object combinations, preferred phrases, sentence complexity, richness of vocabulary and other syntactic measures allows stylometric methods to shine. Finally, political speeches are not complete improvisations. They are drafted by speechwriters and subject to the scrutiny and editing of the speaker themselves. Each speech is thus a narration of a written product, mixed in with the occasional improvisation. It is this primarily written character of modern political speeches that has made stylometry so useful in analysing them.
Any analysis is contingent on having high-quality data in large enough volumes such that the results accurately represent the features of the text being analysed, instead of spurious correlations. In our case, the National Archives of Singapore has complete transcriptions of speeches made by prominent politicians, past and present. Keeping coverage limited to speeches by PAP members also allows us to control for broader ideological differences that might drive word choices made by opposition politicians.
Unfortunately, the extent of coverage differs dramatically. For example, while more than a thousand LKY speeches are available, there is only one by Amrin Amin. This necessitates Amrin’s exclusion due to an insufficient sample size. While there are no hard rules over what constitutes a sufficiently large corpus, I have made my judgements based on the number of available speeches, and their length.
This sample provides adequate coverage across generations and political appointments. We have all four of Singapore’s prime ministers, several prominent Malay and Indian leaders, members from the 4G cabinet as well as women holding ministerial positions.
I transformed their speeches into frequency tables of the most common words and sequences of words used, before allocating a probability of each word or phrase to a speaker. The multi-variable representation of each speaker is then collapsed into two variables using a statistical technique called Principal Components Analysis so that they can be visualised on a two-dimensional plot.
Stylometric Results—LKY: A Man Apart?
The 100 most frequently used words (MFW) include standard ones like “government”, “economy”, “people” and “community”. When we project their relative usage on the plot above, we notice that both LKY and Lawrence Wong occupy distinct positions, indicating they use these terms very differently from all other politicians in our corpus. LKY used them much less, while the current prime minister uses them often. This implies that Lee adhered less rigidly to the bureaucratised talking points in his speeches, which aligns with the notion that he had a unique rhetorical style.
However, individual words alone tell us very little about speeches. We repeat the analysis by paying attention to the most commonly used short phrases of three words (trigrams). These are phrases such as “as a result”, “the development of”, “we need to”, “over the last” and “in the region”, which are once again phrases we would expect to see in boilerplate government orations.
Yet again, LKY stands apart, indicating that he used these common phrases less frequently than other politicians. Wong is more aligned with other speakers this time, while Teo Chee Hean occupies a distinct space in the top right. The differences in results between this plot and the MFW plot could be a product of idiosyncrasies in the words or phrases contained within each speaker’s corpus. For example, if our corpus includes a speech made in response to SARS by one politician but not others, topic-relevant words or phrases would only appear for that speaker, making them unique.
To smooth out these biases in our sample and understand which of our observations are a product of systemic differences in speech instead of one-time quirks arising out of a coincidental choice of words or phrases, we repeat this process over 50 iterations, increasing the number of most frequent words by 100 each time and increasing the length of the phrases being analysed from three to six words to identify common patterns.
The results can be seen in the tree diagram above. The tree branches off based on similarities, two speakers belonging to a single branch are the most similar with the relative positioning of branches denoting similarities between clusters. Names closer to the periphery contain the most distinct corpuses. Across all iterations, LKY continues to occupy a unique position, as do two other former prime ministers, Lee Hsien Loong and Goh Chok Tong.
Comparing prime ministers
This by itself is not definitive, but we now have cause to focus on LKY. In this next section, we’ll cover the three largest corpora in our sample, LKY, Lee Hsien Loong and Goh Chok Tong. We’ll analyse the complexity of their speeches through various syntactic measures using three speeches each made as prime minister, on the same topic; justifying the high salaries for civil servants and ministers.
The first measure is entropy, used in information theory to decompose letters of text into binary bits. Lower entropy values indicate higher structure, and correspondingly, more predictability. In our sample, the mean, median and distribution of entropy values can be visualised with a violin plot.
Interestingly, all three speakers have high entropy. This indicates a lack of structure and high unpredictability associated with speech. The similarity is surprising, hinting that they delivered information to the audience in comparable ways.
To complement entropy, I plot the distribution of polysyllabic words used by each speaker as a measure of speech complexity. LKY stands out as having the highest average and the widest distribution, indicating a greater propensity to use longer and more complex sounding words. In their salary speeches, Lee Hsien Loong and Goh Chok Tong’s only polysyllabic words are those directly pertaining to technocratic aspects of governance such as “development” and “transportation”. Meanwhile, LKY paints with a broader brush, using words such as “discomfiture” and “prerogative” to make rhetorical points contrasting public ire with ministerial pay and his desire to attract “talent” to Singapore politics.
Next, I focus on lexical diversity, which involves analysing the vocabularies of all three speakers via two measures, type-to-token ratio and hypergeometric diversity. Both start by dividing the total number of unique words by the total number of words per speech before performing different transformations to give a measure of lexical diversity.
All three speakers have similarly rich vocabularies, albeit with some notable differences in distribution. LKY uses a mix of simpler and more complex words, while Lee Hsien Loong and Goh Chok Tong tend to stick to more complicated words with more extreme variations than LKY. We can illustrate this difference with snippets from each of their salary speeches.
“Good government is going to remain critical to Singapore in future. This is not just during the first phase—as we take off, you need a strong government; after that, you can fly on autopilot. In fact, I think in the next phase, we will need even more skilled pilots.” - Lee Hsien Loong
“How did we do it? Without wanting to sound immodest, I would say the short answer is: good government. This is not to down-play the role of the people themselves—their diligence, their emphasis on education, their high savings habit, their discipline and above all, their responsiveness to public policies and support for the government. But people, like an orchestra, can play good symphonies only under the hand of a maestro.” - Goh Chok Tong
“So it was an unending quest for the right man to put in the job. It takes years for a person to be tried and tested as a minister, and to develop the judgement and touch. Every cabinet has its core team and core of goal-shooters and goalkeepers. Not all 11 players have the same star quality. It is the prerogative of the prime minister to pick his men, and his responsibility to gel them into a team and get the team to deliver. It is impossible for the PM [prime minister] to pay each player as football clubs do, in accordance to their market price. I would have liked to pay the ministers the way football clubs pay their players. You know, when you have star player and you change clubs, hundreds of millions of pounds change hands. But could I do it? I would cause so much discomfiture, public discomfiture, for the other ministers who are not goal scorers that the team will soon be unable to function as a team. So a certain balance must be kept. Solidarity of the team and recognition by the captain that he is a scorer and he is a first-class goalkeeper, somehow we have got to get them to share not only the glory but also the remuneration.” - LKY
All three of these examples make metaphors about a prime minister and his cabinet, as a pilot, conductor and football player. But only LKY proceeds to build a narrative around the metaphor, in a way that allows it to resonate with the audience, mixing simple words like “goal” and “team”, with more complex ones like “discomfiture”, “prerogative” and “unending”.
Next, I plot the Flesch-Kincaid grade level, a metric that assesses the grade level of the listener required to understand the speech.
Surprisingly, LKY has the tightest distribution of high Flesch-Kincaid scores, requiring on average, approximately 17 years of education to understand, the equivalent of a postgraduate degree. On the contrary, the speeches of Lee Hsien Loong and Goh Chok Tong are more widely accessible, requiring on average between a junior college or university degree to understand. Additionally, their speeches have a wide distribution, indicating a willingness to alter their speeches based on context. This choice, however, might be counterproductive, resulting in an audience feeling like it’s deliberately being talked down to. This is most visible with Lee Hsien Loong’s corpus that includes speeches needing only 0-5 years of education to understand.
This difference is most visible in the national day rally speeches the prime ministers delivered. All three touched on the need to increase the productivity of the Singaporean worker and cautioned against the welfare state model popularised by Germany. But LKY is the only one who mentioned and explained econometric terms such as the “balance of payments deficit” arising from excess government spending whilst Goh Chok Tong and Lee Hsien Loong limited their explanations to “excessive government subsidies” as a cause of economic stagnation. Throughout the various economic examples, LKY provided technical details that might not have been easily understood by the median listener. While these differences are subtle at the individual speech-level, they become obvious in the aggregate.
As a final analytic step to highlight the differences, we perform sentiment analysis on the distribution of speeches by each PM, with a modern language model.
These modern models are bidirectional, performing sequence predictions with information from words before and after the current word. They observe how closely their predictions align with their training data and use that to classify text. In this case, we use a pre-trained sentiment analysis model, trained by Google and used on a large corpus of English text data for maximum accuracy. The distribution results and the magnitude of positive and negative sentiment are displayed below.
Figure 6 displays the distribution of negative sentiment. While all three prime ministers have a similar distribution of negative sentiment, LKY’s speeches have a much higher mean and interquartile range of negative sentiment scores indicating a higher propensity for making negative remarks. Correspondingly, on the green plot for positive sentiment, LKY has a much wider distribution of positive sentiment scores in the low ranges with a smaller mean. He thus conveys the most varied and negative sentiment in his speeches. Lee Hsien Loong and Goh Chok Tong on the other hand, have wide distributions of negative sections, indicating less propensity to say something negative and a tighter distribution of text selections with positive sentiment.
In a sense, this reflects the historical circumstances and realities in which each of these leaders functioned. LKY consistently described himself and his original cabinet as belonging to a unique generation.
“As I have said, the first cabinet consisted of men who had been drawn together by the forces of history. We were young men in our 30s and early 40s, who had been through the sufferings of the war and the Japanese Occupation. We were tempered by that experience to seek independence, strongly driven to change the nature of our society and our economy.”
This paragraph lends him the ability to criticise Singaporeans in the same speech without sounding arrogant; “one weakness of the democratic system is that voters do not have a proper assessment of the real person behind a first-time candidate”.
Goh Chok Tong and Lee Hsien Loong, by contrast, routinely pay homage to LKY’s generation but refrain from paternalistically chiding their electorate. Lee Hsien Loong for example, makes the same case albeit differently. “I hope that when we go to elections, we will have a good record and the voters will assess the government fairly, based on overall results for the whole term. That is how a democratic system has to work, and to work well, it depends on the government performing well and it depends on the electorate making the right judgments.”
Once again, these are subtle differences but they show up more distinctly in the aggregate.
The results demonstrate that LKY has a unique corpus of speeches in terms of its most frequent words used, combinations of words used, complexity, lexical diversity, and sentiment. His propensity for negative sentiment is particularly notable, tying in with his image as a no-nonsense leader, always willing to dish out the hard truths to Singaporeans.
When I ask my dad why he enjoys listening to LKY and not other leaders, he always reiterates the same point. Lee senior had the ability to project strength, making his successors look weak and tame by comparison. This strongman image is not just created by the state’s historical narratives. It is visible in the way Lee delivered speeches in statistical aggregate. A distinct vocabulary and tone, set him apart from other politicians.
Another important takeaway from this analysis is how visibly Singapore’s perceptions of current political discourse is shaped by nostalgia. We view our older leaders as superior even though most of our current leaders have emerged from the selection process our founding fathers devised. They are well credentialed, possess clearly defined meritocratic achievements, and are elected with a commitment to technocratic efficiency while avoiding naïve idealism, exactly as their predecessor intended. To be unhappy with current leaders would be to be unhappy with the products of our underlying social contract rather than any specific breaks with the past.
Perhaps our current leaders need to create their own unique styles, conceived of independently and without passing through the multiple layers of public relations vetting and speechwriting common in modern politicking. And perhaps, it is true that despite our most idealistic aspirations for an informed public, voters still secretly crave the harsh put-downs of Lee’s era. It remains to be seen how far this psyche will matter in the upcoming GE but it is certainly not a “hard truth” that our 4G leaders can afford to ignore.
Glossary
Bootstrap Consensus Tree: A visualisation method that shows relationships between texts or authors by combining multiple bootstrap samples of stylometric features, displaying only the groupings that consistently appear across resampled analyses, thus providing a robust representation of stylistic similarities.
Entropy: A measure of the unpredictability or randomness in text, calculated using the probability distribution of characters or words, where higher entropy indicates greater uncertainty and complexity in the text’s content. In information theory, it represents the average amount of information contained in each symbol or word.
Flesch-Kincaid Scores: Readability metrics that measure text complexity based on word and sentence length, with the Grade Level score indicating the US grade level required to understand the text, and the Reading Ease score ranging from 0-100 (higher scores indicating easier readability).
Hypergeometric Diversity Score: A statistical measure that evaluates how distinctively frequent a word is in a specific text compared to its overall frequency in a larger corpus, using the hypergeometric probability distribution to identify words that appear more often than would be expected by chance.
Most Frequent Words (MFW): A text analysis metric that identifies and ranks words by their occurrence count in a document or corpus, often used in stylometry after removing stopwords to reveal characteristic vocabulary patterns of an author or text.
Ngrams: Sequences of N-consecutive elements (usually letters, characters, or words) in a text, used to analyse patterns in language use, authorship style, or to build language models.
Principal Component Analysis (PCA): A dimensionality reduction technique that transforms high-dimensional data into a new set of uncorrelated variables (principal components) ordered by importance, with each component capturing the maximum possible variance in the data while being orthogonal to previous components.
Sentiment Analysis: A natural language processing technique that determines the emotional tone or opinion (positive, negative, or neutral) expressed in text by analysing word choice, context, and linguistic patterns.
Stylometry: A method of analysing linguistic style through statistical and linguistic features (like word choice, sentence length, punctuation patterns) to derive writing patterns.
Type-to-token ratio (TTR): A measure of lexical diversity calculated by dividing the number of unique words (types) by the total number of words (tokens) in a text, resulting in a value between 0 and 1, where higher values indicate greater vocabulary richness.
Thiyaghessan Poongundranar is a Singaporean based in Seattle. He works in public policy, with a focus on the Social Sector in the United States.
Letters in response to this piece can be sent to sudhir@jom.media. All will be considered for publication on our “Letters to the editor” page.
