Is Trump's Vocabulary Poor? Vocabulary Richness Across Texts of Different Lenghts
Dominique Labbe, Cyril Labbe, Jacques Savoy
cs.CL
Quick verdict
Claims Trump's debate vocabulary is the poorest since 1960 using a vocabulary partition model; transparent but lacks benchmarks and multiplicity control.
Summary
The paper asks whether Donald Trump's vocabulary in US presidential debates is poorer than his opponents'. Using 68 debate transcripts (1960-2024, 19 nominees, 34 debates) from the American Presidency Project, the authors argue that classic richness measures (TTR, Honore's H, MATTR, big-word percentage, lexical density) are unstable across text lengths, and propose a 'vocabulary partition model': vocabulary growth is split into a general component following Muller's hypergeometric urn formula plus a specialized component growing linearly with slope p, fitted by least squares. The theoretical vocabulary of a longer text at a shorter text's length comes with a variance-derived fluctuation interval (alpha = 5%/1%), enabling significance assessment between unequal-length speeches. Results: Trump's seven debates (2016-2024) fall within one narrow diversity band; against Biden (first 2024 debate) his expected vocabulary is ~11% lower; his richest speech (2020, second debate) is statistically indistinguishable only from Nixon's first 1960 debate and Romney's 2012 debates, and significantly poorer than all other nominees since 1960. The authors conclude Trump's vocabulary is uniquely poor, reflecting communication strategy or personality. Main limitations: no benchmarking against established lexical-diversity measures, no multiple-comparison control, fitted-p uncertainty apparently not propagated into intervals, and unexamined confounds from 60 years of debate-format and topical drift.
Review complete
Assessment at a glance
6
risk flags identified
● high 0
● medium 4
● low 2
4 strengths · 5 weaknesses · 5 author questions
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Vocabulary growth in natural texts is better explained by a two-component partition model (general hypergeometric reservoir plus linearly growing specialized component) than by Muller's single urn model, which overestimates by more than 10% mid-text.
- Trump's vocabulary richness in all seven 2016-2024 debates is statistically stable and significantly lower than Hillary Clinton's, Biden's, and Harris's (about 11% below Biden in the first 2024 debate).
- Against all debate participants since 1960, only Nixon's first 1960 debate and Romney's first two 2012 debates are statistically indistinguishable from Trump's best performance; all others are significantly richer.
- Trump therefore has the poorest debate vocabulary of the 19 nominees in the corpus, with less than 1% risk of error for the key comparisons.
Strengths
Well-controlled comparison setting
Section 3 motivates debates as a near-laboratory design: equal speaking time, similar format, oral mode ruling out ghostwriters; transcription normalization is specified (case folding with US vs. us kept distinct, number and dollar-sign standardization).
Explicit significance machinery for unequal-length texts
Equations 3-8 convert vocabulary-growth curves into comparable theoretical values with 5%/1% fluctuation intervals, directly addressing the length-dependence of TTR and Honore's H that the authors ground in Baayen's LNRE work (Section 2).
Model demonstrably improves fit over the Muller baseline
Exhaustive, verifiable numeric reporting
Weaknesses
No benchmark against established measures
Statistical machinery under-specified and possibly optimistic
General/specialized partition lacks linguistic validation
Sixty years of format drift left unexamined
Normative framing outruns the measure
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
- All 4 strengths & 5 weaknesses, analyzed in detail
- 6 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
Does this study prove Trump has a poor vocabulary?
Not in a general sense. The authors show that across 68 US presidential debate transcripts (1960-2024), the number of distinct words Trump used - statistically adjusted for speech length via their partition model - is significantly lower than nearly all opponents'; his richest debate matches only Nixon's first 1960 debate and Romney's 2012 debates. The measure captures type diversity in debate settings, not overall language ability or intelligence.
How can vocabulary richness be compared fairly between speeches of different lengths?
The paper models vocabulary growth as a general component following a hypergeometric urn law plus a linear 'specialized' component. This predicts how many distinct words a longer speech would contain if cut to a shorter speech's length, with confidence intervals; if the shorter speech's observed vocabulary falls outside that interval, the difference is declared statistically significant.
What data did the researchers analyze?
Sixty-eight transcripts covering 34 US presidential general-election debates from 1960 to 2024, involving 19 nominees, downloaded from the American Presidency Project (presidency.ucsb.edu) and processed with standardized tokenization rules.
Which candidates had the richest debate vocabulary in this study?
Among 2016-2024 participants, Hillary Clinton (all three 2016 debates) and Harris (2024) rank highest, Biden sits in a middle band, and Trump ranks lowest. Extending back to 1960, nearly all nominees - including Obama, Kerry, Gore, and both Bushes - significantly exceed Trump's best performance.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 1-5 are not analyzable from the extracted text; fit-quality judgments rely on the authors' descriptions.
- Several equations (notably Eq. 6 for p and Eq. 8 for the variance) are garbled in extraction, so exact computational details of the variance could not be verified.
- Venue metadata is a placeholder ('Glottometrics XX, 20XX'); this review covers the preprint as extracted.
- The title typo 'Lenghts' is reproduced from the source document.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.