Does Donald Trump speak differently from other modern American presidents? A new study in PNAS nexus suggests that he does. By applying machine learning to a wide range of speeches—from debates to campaign rallies and State of the Union addresses—the team found that Trump’s rhetoric is distinguished by short, direct sentences and a noticeably hostile tone, especially toward opponents. The study’s AI-driven analysis highlights Trump’s unique style, which sets him apart from other presidents, both Republican and Democratic.
The research is inspired by questions about whether presidents speak in ways that can be noticeably unique and how those differences might reveal themselves in different speaking contexts. A presidential address is one of the most direct ways a leader connects with the public and influences perceptions of policies and opponents. The researchers wanted to see if the language of modern presidents could be analyzed quantitatively to reveal different styles, and whether those styles could indicate different levels of unity or division in their message.
“We are interested in how large language models can help us understand and quantify presidential discourse. One of the striking phenomena is the uniqueness of Trump’s speeches, and this work allows us to answer this question quantitatively,” explains study author Chenhao Tan, assistant professor at the University of Chicago and director of the Chicago Human+ AI lab.
To study this, the researchers analyzed a range of texts, including 35 transcripts of presidential debates, 67 State of the Union addresses, and 187 campaign speeches by presidents since 1960. By examining these different types of speeches, they ensured that the style of each president was assessed. both in structured, formal contexts and in more flexible, high-stakes debate settings.
A central aspect of the research was the creation of a new ‘uniqueness’ metric, developed using a machine learning approach called large language modelling, which identifies patterns in language use. For this metric, the researchers trained language models on specific data sets for each type of speech (debates, State of the Union addresses, and campaign events), allowing the model to detect how likely a given sentence was to be unique to one president compared to others.
Essentially, this metric captured the predictability of certain words and phrases based on the speaker. The greater the variation from typical word patterns, the higher a speaker’s uniqueness score. For example, if President Trump used language structures or expressions that other presidents were unlikely to use, his uniqueness score would be higher, as was often the case in the findings.
The researchers also introduced a divisive lexicon, a list of 178 words commonly used to insult, discredit or delegitimize opponents. This list was developed from an initial selection of divisive words, which were then reviewed by multiple researchers to confirm their relevance to political speech. Words like “corrupt,” “stupid,” and “disgrace” were included in this lexicon, which was then applied to each presidential data set to measure the frequency of divisive language.
The researchers found that Trump’s language was both highly unique and divisive, setting him apart from other modern presidents in important ways. Trump’s use of short, direct sentences was consistent across all categories of speeches examined: debates, campaign speeches, and State of the Union addresses. In contrast to the more complex or measured phrasing typical of many other presidents, Trump displayed more simplistic and repetitive language.
One striking observation was that Trump’s speech patterns were not only different from those of Democratic presidents, but also significantly different from those of his fellow Republicans. This finding was unexpected; Traditionally, within-party presidential speech exhibits relatively minor stylistic differences compared to cross-party language.
However, the analysis found that Trump’s language style within his own party was a bigger outlier than the differences typically observed between Republicans and Democrats overall. This suggests that Trump’s rhetorical approach represents a unique type of discourse, deviating from both standard Republican and Democratic norms.
“We expected Trump not to stand out in State of the Union speeches as these speeches are more standard,” Tan told PsyPost. “However, he is still very different in these speeches. I was also surprised by how different Trump is from other Republicans.”
Trump also mentioned his opponents more often than his predecessors. But it wasn’t just the amount of these references that stood out; Trump’s language also included unique descriptions when talking about opponents. This uniqueness was demonstrated by a lower degree of overlap between the adjectives Trump used and the adjectives typically used by other presidents to describe their rivals.
While other presidents might have relied on more neutral or policy-oriented terms, Trump often used emotionally charged descriptors that cast his opponents in a particularly negative light. This approach made his language about opponents distinctly personal and combative, emphasizing a style that separated him from both Republican and Democratic predecessors.
“Donald Trump’s political rhetoric is unique among modern presidents and is defined in part by his use of antagonistic language, especially when directed at political opponents,” Tan said.
While the findings provide insight into Trump’s rhetorical style, the researchers acknowledged some limitations in their approach. For example, campaign speech dates only include presidents from recent election cycles, limiting comparisons with past presidents. Furthermore, the study focused exclusively on public speeches, excluding social media posts, which play an important role in modern political communication.
“We couldn’t get much high-quality data for campaign speeches, especially for early presidents,” Tan noted. “I think our divisive lexicon, while new, may need more iterations to ensure that it accurately reflects the divisive nature of a speech. The nice thing is that these approaches, based on word counting, are interpretable.”
Despite these limitations, the study provides a comprehensive view of the unique nature of Trump’s speech and how it differs from those of other presidents, underscoring a shift toward a more confrontational style in presidential discourse.
“We hope to help the public make informed decisions in democratic processes,” Tan said. “Actually, we just built a website to show some sample summaries about political platforms: https://chicagohai.github.io/election-nlp/.”
The study, ‘Quantifying the uniqueness and divisiveness of presidential discourse’, was authored by Karen Zhou, Alexander A Meitus, Milo Chase, Grace Wang, Anne Mykland, William Howell and Chenhao Tan.
