The Rhetorical Crucible
Word patterns in three presidential debates
The Rhetorical Crucible
A word can sound different depending on who says it, what comes before it, and when it appears. Choose a word from the debate transcripts and follow it through three views: isolated uses, surrounding context, and a timeline of appearances. The visual patterns invite a closer reading of the transcript; they do not reveal a speaker’s intentions.
Each candidate’s words are analyzed separately. The animation groups words used at similar points in the transcript; the record below shows the counts and reasons for each grouping.
The charge enters in public
The Word Trails galley retains chronology: a word’s height is where it peaked in the transcript, its size is that peak-bin count, and its hue records which voices used it. The brass word is the one followed through both candidate retorts below.
Two retorts, one clock
Candidate vocabularies are processed independently against the same five-part chronology. Near-matching 20-bin profiles open an alloy; later words may join only when their mean similarity clears the declared threshold.
leaves in JOBS + WORK + ECONOMY
Opened JOBS + WORK + ECONOMY at cosine 0.81.
The same word, used at different points in the debate.
leaves in COMPANIES + JOBS
Opened COMPANIES + JOBS at cosine 0.94.
Words belong together here when their counts rise and fall at similar points in the transcript. Return to those passages to examine what the speaker was saying.
What survived the heat
The names below are deliberately unromantic: they contain only the words that formed them. No topic label, ideological diagnosis, or victory was smuggled in by the animation.
PEOPLE + NEED + WILL
people · need · will · country
4 words · 93 usesJOBS + WORK + ECONOMY
jobs · work · economy
3 words · 45 usesCOMPANIES + JOBS
companies · jobs
2 words · 39 usesCOUNTRY + SECRETARY + CLINTON
country · secretary · clinton · good
4 words · 120 usesMONEY + GREAT
money · great
2 words · 33 usesOpen the complete disposition ledger · 24 admitted words
| Candidate | Rank | Word | Uses | Disposition | Destination | Recorded basis |
|---|---|---|---|---|---|---|
| Clinton | 1 | people | 33 | alloyed | PEOPLE + NEED + WILL | Joined PEOPLE + NEED + WILL at mean cosine 0.69. |
| Clinton | 2 | donald | 26 | unalloyed | UNALLOYED | No unassigned partner met the 0.65 opening-pair threshold. |
| Clinton | 3 | need | 23 | alloyed | PEOPLE + NEED + WILL | Opened PEOPLE + NEED + WILL at cosine 0.83. |
| Clinton | 4 | will | 21 | alloyed | PEOPLE + NEED + WILL | Joined PEOPLE + NEED + WILL at mean cosine 0.61. |
| Clinton | 5 | good | 17 | unalloyed | UNALLOYED | No unassigned partner met the 0.65 opening-pair threshold. |
| Clinton | 6 | jobs | 17 | alloyed | JOBS + WORK + ECONOMY | Opened JOBS + WORK + ECONOMY at cosine 0.81. |
| Clinton | 7 | country | 16 | alloyed | PEOPLE + NEED + WILL | Opened PEOPLE + NEED + WILL at cosine 0.83. |
| Clinton | 8 | tax | 16 | unalloyed | UNALLOYED | No unassigned partner met the 0.65 opening-pair threshold. |
| Clinton | 9 | business | 15 | unalloyed | UNALLOYED | No unassigned partner met the 0.65 opening-pair threshold. |
| Clinton | 10 | work | 15 | alloyed | JOBS + WORK + ECONOMY | Joined JOBS + WORK + ECONOMY at mean cosine 0.67. |
| Clinton | 11 | american | 13 | unalloyed | UNALLOYED | No unassigned partner met the 0.65 opening-pair threshold. |
| Clinton | 12 | economy | 13 | alloyed | JOBS + WORK + ECONOMY | Opened JOBS + WORK + ECONOMY at cosine 0.81. |
| Trump | 1 | country | 51 | alloyed | COUNTRY + SECRETARY + CLINTON | Joined COUNTRY + SECRETARY + CLINTON at mean cosine 0.71. |
| Trump | 2 | people | 37 | unalloyed | UNALLOYED | The three-product limit closed before another qualifying opening pair could be formed. |
| Trump | 3 | will | 32 | unalloyed | UNALLOYED | The three-product limit closed before another qualifying opening pair could be formed. |
| Trump | 4 | secretary | 27 | alloyed | COUNTRY + SECRETARY + CLINTON | Opened COUNTRY + SECRETARY + CLINTON at cosine 0.92. |
| Trump | 5 | years | 24 | unalloyed | UNALLOYED | The three-product limit closed before another qualifying opening pair could be formed. |
| Trump | 6 | clinton | 22 | alloyed | COUNTRY + SECRETARY + CLINTON | Opened COUNTRY + SECRETARY + CLINTON at cosine 0.92. |
| Trump | 7 | companies | 20 | alloyed | COMPANIES + JOBS | Opened COMPANIES + JOBS at cosine 0.94. |
| Trump | 8 | good | 20 | alloyed | COUNTRY + SECRETARY + CLINTON | Joined COUNTRY + SECRETARY + CLINTON at mean cosine 0.63. |
| Trump | 9 | jobs | 19 | alloyed | COMPANIES + JOBS | Opened COMPANIES + JOBS at cosine 0.94. |
| Trump | 10 | money | 17 | alloyed | MONEY + GREAT | Opened MONEY + GREAT at cosine 0.85. |
| Trump | 11 | bad | 16 | unalloyed | UNALLOYED | The three-product limit closed before another qualifying opening pair could be formed. |
| Trump | 12 | great | 16 | alloyed | MONEY + GREAT | Opened MONEY + GREAT at cosine 0.85. |
The assayer’s fine print
A metaphor earns its keep by clarifying the rules under which it may be trusted. These are the rules, including the rather important ones that keep the furnace from manufacturing its own evidence.
The transcript is tokenized, common stopwords are removed, and each speaker’s remaining words are counted in twenty equal token-position bins. These are positions in a transcript, not minutes on a clock. The twelve most-used non-artifact words for each candidate constitute the admitted charge.
Every admitted pair is compared by cosine similarity across all twenty bins. An unassigned pair scoring at least 0.65 may open an alloy. At most two more words may join when their mean similarity to its present members is at least 0.55. Each source word belongs to one destination only; no word is quietly duplicated for dramatic effect.
Each candidate can form up to three groups. Words that do not qualify stay separate, with the reason recorded. Transcript artifacts such as “crosstalk” and “nbsp” are removed before the analysis.
Choosing a debate runs the same analysis on that transcript’s word counts. The chart animates the resulting assignments so you can follow how the groups form.
1 transcript-artifact profiles
nbsp
Twenty-four admitted sources
Twelve per candidate; each closes in one alloy or one unalloyed outlet. The amount ledger remains the original word-use count.
Timing, not meaning
Cosine similarity compares the complete 20-bin timing profiles; it does not infer topic, intent, agreement, or debate performance.