R also has a wordcloud package, though it’s hard to beat Wordle on looks.
# Load raw data, stored at textuploader.com speech.raw
Calculating textual metrics
library(qdap) library(data.table) # Split into sentences # qdap's sentSplit is modeled after dialogue data, so person field is needed speech.df
with(sentences, plot(pct.complete, pol))
Cleaning up the plot and adding a LOESS smoother clarifies this trend, particularly the peak at the end.
library(ggplot2) library(scales) my.theme
sentences[, readability := automated_readability_index(speech, sentence.num) $Automated_Readability_Index]
CustomScatterPlot(ggplot(sentences, aes(pct.complete, readability)) + ylab("Automated Readability Index") + ggtitle("Readability of I Have a Dream speech"))
Scraping Google search hits
sentences[, google.hits := GoogleHits(paste0("[", gsub("[,;!.]", "", speech), "] mlk"))]
ggplot(sentences, aes(pct.complete, google.hits / 1e6)) + geom_line(color="grey40") + # Lighten dots xlab("Percent complete (by syllable count)") + scale_x_continuous(labels = percent) + my.theme + ylim(0, max(sentences$google.hits) / 1e6) + ylab("Sentence memorability (millions of Google hits)") + ggtitle("Memorability of I Have a Dream speech")
 "free at last!"  "I have a dream today."  "I have a dream today."  "This is our hope."  "And if America is to be a great nation this must become true."  "I say to you today, my friends, so even though we face the difficulties of today and tomorrow, I still have a dream."  "We cannot turn back."
Plotting Google hits on a log scale reduces skew and allows us to work on a ratio scale.
sentences[, log.google.hits := log(google.hits)] CustomScatterPlot(ggplot(sentences, aes(pct.complete, log.google.hits)) + ylab("Memorability (log of sentence's Google hits)") + ggtitle("Memorability of I Have a Dream speech"))
What makes a passage memorable? A linear regression approach
library(MASS) # For stepAIC google.lm
Call: lm(formula = log(google.hits) ~ poly(readability, 3) + pct.complete.100, data = sentences) Residuals: Min 1Q Median 3Q Max -4.2805 -1.1324 -0.3129 1.1361 6.6748 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 11.444037 0.405247 28.240 < 2e-16 *** poly(readability, 3)1 -12.670641 1.729159 -7.328 1.75e-10 *** poly(readability, 3)2 8.187941 1.834658 4.463 2.65e-05 *** poly(readability, 3)3 -5.681114 1.730662 -3.283 0.00153 ** pct.complete.100 0.013366 0.006848 1.952 0.05449 . --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 1.729 on 79 degrees of freedom Multiple R-squared: 0.5564, Adjusted R-squared: 0.534 F-statistic: 24.78 on 4 and 79 DF, p-value: 2.605e-13
The effect of pct.complete can be calculated by exponentiating the coefficient, since I log-transformed the outcome variable:
exp(google.lm$coefficients["pct.complete.100"]) pct.complete.100 1.013456
This result can be interpreted as the following: a 1% increase in the location of a sentence in the speech was associated with a 1.3% increase in search hits.
Interpreting the effect of readability is not as straightforward, since I included polynomials. Rather than compute an average effect, I graphed predicted Google hits for values of readability's observed range, holding pct.complete.100 at its mean.
This cubic relationship indicates that predicted memorability falls considerably until about grade level 10, at which point it levels off (very few passages have readability exceeding 25).
- The speech starts and (especially) ends on a positive note, with a positive middle section filled with two troughs to vary the tone.
- While readability/complexity varies considerably within each small section, the overall level is fairly consistent throughout the speech.
- Readability and placement were the strongest drivers of memorability (as quantified by Google hits): sentences below grade level 10 were more memorable, as were those occurring later in the speech.
To a degree, these were intuitive findings--the ebb and flow of intensity and sentiment is a powerful rhetorical device. While we may never be able to fully deconstruct the meaning of this speech, techniques explored here can provide brief insight into the genius of MLK and the power of his message.
Thanks for reading, and enjoy your MLK day!
- Special thanks to Ben Ogorek for guidance on some of the statistics here, and for a thorough review.
- Special thanks to Mindy Greenberg for reviewing and always pushing my boundaries of conciseness and clarity.
- Thanks to Josh Kraut for offering a ggplot2 lesson at work, inspiring me to use it here.