Source-linked AI summary
Evolution of US Oral Political Language
Jacques Savoy
TL;DR
Prepared written messages may not reflect politicians’ personal oral style, motivating analysis of US presidential-debate speech. Across decades, the study finds declining language sophistication and shorter sentences, alongside a shift toward more emotional and less logical expression.
Problem
Prior analyses often use prepared written political messages, which may not reflect a politician’s personal style and rhetoric, leaving oral political language less directly examined.
Source: §1 · p. 1Method
The study analyzes presidential-debate transcripts to assess candidates’ oral language and style over time.
Source: §1 · pp. 1–2Results
Across decades, language sophistication declines, and recent nominees average 11 tokens per sentence versus 22.3 tokens for older candidates, with recent candidates using more emotional and less logical expression.
Source: §6 · pp. 12–13Takeaways & Limitations
Trump’s language follows broader trends toward repetition, simpler terms, and shorter sentences also seen in contemporary candidates such as Biden.
Source: §6 · p. 13Measures of vocabulary richness are not stable across texts of different lengths from the same source, limiting their use as distinctive stylistic features.
Source: §4 · p. 7
Abstract
from arXiv · showhide
The analysis of US political language is usually based on the written form (e.g. presidential addresses) or posts broadcasted on various social networks. Oral production, however, which is even more frequent, can better reveal the style and mode of thinking of the speaker. This study covers this mode of linguistic communication by considering 19 candidates from the presidential elections between 1960 to 2024. Our main research objectives are to disclose the main trends hidden in those presidential debates. Do we observe a clear simplification of the US political language over time? Does Trump have poor language compared to the other candidates? Do unusual stylistic features occur only with a single, specific president? Moreover, can we detect a pattern explaining the success or failure of some nominees? Over time, this study demonstrates a significant reduction in political language complexity, a decrease of the mean sentence length, and a noteworthy decline of complex terms. Moreover, the emotional tone increases over the decades, while logical and rational thinking tends to lessen.
1 Introduction
Prior analyses often rely on prepared written political messages, which may reflect collaboration with staff rather than a politician’s personal style. This study turns to presidential-debate speech to examine change over time, candidate differences, and possible stylistic patterns linked to electoral success or failure.
1 Introduction
- Written political messages may reflect collaboration between politicians and staffers, so they may not capture a politician’s personal style and rhetoric.The study instead uses presidential-debate transcripts to examine oral production as a way to assess candidates’ style and thinking mode.Source: §1 · p. 1
- The study asks whether US oral political language has changed over more than 60 years, whether trends recur across candidates, and whether Trump is an outlier.It also asks whether debate styles reveal features associated with nominees’ success or failure.Source: §1 · p. 2
- The analysis examines vocabulary complexity, language sophistication, sentence length, and selected word-frequency measures.The introduction frames persuasion as potentially relying on rational argument or simpler, more emotionally intense expression.Source: §1 · p. 2
- Oral speech differs from written or web-based language through transience, spontaneity, shorter sentences, repetition, fillers, and possible interruption.The debate transcripts omit intonation, volume, tempo, gestures, and facial expressions.Source: §1 · p. 2
2 Related Work
Related work studies political language through written speeches, web-based communication, vocabulary and syntax measures, and rhetorical or psychological word categories. The review also notes limits in automated interpretation of tone and psychological traits.
2 Related Work
- Political-style studies commonly analyze selected written messages, including State of the Union and end-of-year speeches, which do not closely represent oral communication.Prior work also analyzes congressional and presidential speeches, including Lim’s five trends in more recent discourse: antiintellectualism, poetic terms, assertive tone, people-oriented rhetoric, and colloquial language.Source: §2 · pp. 2–4
- Vocabulary richness studies use measures such as Type Token Ratio, hapax density, and the percentage of words with six or more letters.Syntax studies also measure mean sentence length; long sentences can indicate detailed reasoning but may impede easy understanding (Labbé & Savoy 2020).Source: §2 · p. 3
- Web-based political posts can use short formulations, slang, emoticons, emojis, and hyperlinks to attract attention or support claims (Savoy 2018).They let politicians avoid traditional media and press conferences.Source: §2 · p. 3
- Rhetorical wordlists group terms into categories such as Symbolism, Blame, and Communication, while LIWC groups words into syntactic, emotional, or psychological classes.Hart (1984) and Hart et al. (2013) use Diction to compare presidential rhetoric; Slatcher et al. (2007) use LIWC measures to profile 2004 candidates.Source: §2 · pp. 3–4
- Topic-model analyses report a shift in State of the Union topics from statecraft and political economy before 1917 toward domestic and foreign policy afterward (Rule et al. 2015).Automated extraction of political tone or psychological traits also faces concerns including lexical ambiguity, short context, idioms, and varying term intensity.Source: §2 · p. 4
3 Corpus
The corpus contains presidential-debate transcripts from 19 nominees across 1960–2024, with incomplete debate coverage before 1988 and no debates from 1964 to 1972. The section reports shorter speech turns among later nominees, while noting interruptions as a possible partial explanation.
3 Corpus
- The corpus covers 34 debates and 19 nominees from 1960 to 2024, but complete coverage begins in 1988 and no debates occurred from 1964 to 1972.Table 1 summarizes debate counts, transcript length, and speech-turn statistics.Source: §3 · p. 4, Table 1 · p. 5
- Debates are moderated discussions with candidate responses to topics and questions; the format can allow comparisons under similar questions, speaking time, and setting.The moderator may also give candidates the floor directly.Source: §3 · pp. 4–5
- The corpus contains 455,695 tokens, averaging 16,878 tokens per candidate; Clinton’s 1992 campaign has the largest single-candidate total at 30,553 tokens.Trump has the highest total across his three campaigns and seven debates, at 54,920 tokens.Source: §3 · p. 5
- 303 words versus 76.5 words: mean answer length falls from Nixon–Reagan nominees to Obama–Trump nominees, then rises to 176.7 in 2024.The figure tracks the median number of words per speech turn; short replies of six words or fewer were excluded from the mean comparison.Source: Figure 1 · p. 6, §3 · p. 6
- Examples contrast Dukakis’s 1988 and Obama’s 2008 objections (“That's not true. That's not true.”; “That's not true, John”) with Trump’s 2016 interruptions, simplified and more frequent (“Wrong. Wrong”), which could partly explain reduced speech turns.Jim Lehrer moderated 12 debates from 1988 to 2012, so the moderator cannot be the main factor explaining the decrease.Source: §3 · p. 6
4 Vocabulary Complexity and Sentence Length
The section measures lexical complexity and sentence length, finding a broad shift toward simpler oral political language across nominees over time.
4 Vocabulary Complexity and Sentence Length
- TTR divides vocabulary size by text length, but it is not stable across different-length texts from the same source; MATTR instead uses a moving average.The moving-average computation is presented as a way to address that drawback.Source: §4 · p. 7
- Trump and Biden had significantly below-average MATTR, while Harris was within the mean range; Harris's MATTR was 13% above Trump's and her MSL 53% higher.The paper suggests these differences in oral communication form could partially explain Harris's failure or difficulty being clearly understood.Source: §4 · pp. 7, 9
- MATTR averages 0.404 through Clinton versus 0.362 for the last three nominees, while big-word share and mean sentence length also decline over time.The table presents MATTR, big-word percentage, and mean sentence length; the reported comparisons indicate simplification across these measures.Source: §4 · pp. 7–9, Table 2 · p. 8
- 28.1% (Carter, 1980) versus 15.3% (Trump, 2020) marks the highest and lowest big-word shares, with later nominees favoring shorter wording.Big words are defined as words of six letters or more.Source: §4 · pp. 7–8
- 10 tokens/sentence (Trump, 2020) was the minimum, versus Carter's 23.9 tokens/sentence (1980); recent candidates averaged 12, compared with 22.3 from Nixon to Reagan.The difference between recent and older candidates' mean sentence lengths was significant in a t-test.Source: §4 · p. 9
5 Emotional and Rational Language
The section uses LIWC composite measures to examine emotional tone, confidence, analytical thinking, and authenticity in debate language. Results contrast candidates and show changes across eras and, for emotional tone, across campaigns.
Measures
- The analysis uses four standardized LIWC composites—Tone, Confidence (Clout), Analytical thinking, and Authenticity—with scores from 1 to 100.Tone combines positive and negative emotion categories; the measures use LIWC word categories, and Table 3 reports candidate-level results with statistically significant differences marked by †.Source: §5 · pp. 9–10
Emotional Tone
- Trump’s emotional Tone score was 32.06, the lowest, while G. Bush’s 67.30 was the highest; Biden was second-lowest at 39.94.Scores below 50 indicate generally negative sentiment; values above 50 indicate an overall positive tone.Source: §5 · p. 10
Confidence (Clout)
- Confidence scores were low for candidates from Nixon to Reagan, while the last four candidates scored high; H. Bush had the lowest score, 63.05.H. Bush used I-words at 4.69%, compared with Biden’s 2.07%.Source: §5 · pp. 10–11
Analytical thinking
- Trump’s Analytical thinking score was 26.07, the lowest; Ford scored 84.20, followed by Carter at 79.31 and Kennedy at 74.57.Dole had the second-lowest score, 44.02; the section describes recent candidates as relying more on emotional language and less on logical reasoning.Source: §5 · p. 11
Authenticity
- Gore had the highest Authenticity value at 46.18, followed by H. Bush at 42.29 and Dole at 39.27; Biden was lowest at 17.17.H. Clinton scored 21.63 and Harris 22.02; Trump was described as closer to the people than his opponents.Source: §5 · p. 12
Trends across candidates and campaigns
- Recent candidates were described as less optimistic, more certain, less rational, and more authentic and people-oriented than nominees from Nixon to Reagan.Emotional Tone also varied across campaigns: it fell for both H. Bush and G. Bush between their first and second campaigns, while Obama’s rose from 55.01 in 2008 to 63.55 in 2012; Trump’s lowest campaign value was in 2024.Source: §5 · p. 12
6 Conclusion
The conclusion reports simplification in oral political language over time, alongside shifts in emotional and psychological measures. It argues that Trump extends several contemporary language trends rather than standing apart from them.
Language sophistication
- Political language sophistication significantly decreased over decades, with higher lexical repetition and lower Big Words scores among contemporary candidates, including Biden and Trump.The authors say contemporary candidates lean toward shorter words that are easier for audiences to understand.Source: §6 · p. 12
Sentence length
- Recent nominees averaged 11 tokens per sentence, compared with 22.3 for older candidates.The conclusion links shorter sentences with easier understanding and says substantiated reasoning or detailed explanation was more typical in the past.Source: §6 · p. 12
Emotional tone and confidence
- Nominees were positive in mean emotional Tone, but the last two presidents adopted a more negative viewpoint; candidate confidence increased over the years.The conclusion also describes recent candidates as more certain in their wording and conviction.Source: §6 · pp. 12–13
Analytical thinking
- The conclusion describes a shift from Nixon-to-Carter logical, rational expression toward more emotional language and less logical reasoning among H. Clinton, Trump, and Biden.It notes that negative emotions such as anger and fear had already appeared in political advertisements, whose effectiveness depends on tone, party, and topics.Source: §6 · p. 13
Trump in context
- Trump is not a significant outlier: his repetitive wording, avoidance of complex terms, and short sentences continue trends also seen in Biden, though his negative tone and authenticity distinguish him.The conclusion says Trump favors expressions likely to attract audience interest and emerge as media headlines.Source: §6 · p. 13
Appendix
The appendix contains two tables of LIWC results: one organized by nominee and year, and another by candidate and year.
Table A.1
- Table A.1 presents results for selected LIWC categories by nominee and year.Source: Table A.1 · p. 14
Table A.2
- Table A.2 presents LIWC composite summary measurements by candidate and year.Source: Table A.2 · p. 15