The purpose of this post is twofold: (1) to introduce rperseus, my latest R package; and (2) to venture a sentiment analysis of the four gospels.
Rivers of ink have been spilled over the unity and disunity of the four gospels. Their intertextualities have inspired almost two millenia of
speculation both scholarly and pious. We may never know if Q existed, or discern John’s stages of composition, or figure
out what the devil bdelugma tes eremoseos (NRSV: “desolating sacriliege”) means. But now, almost 2000 years removed from their
original composition, we can do a sentiment analysis! And we owe it all to tidytext and the
good people over at the Perseus Digital Library.
Getting the Text
First, we need the gospel texts. For mortals this is impossible, but with R all things are possible1. Towards that end, rperseus combines seamlessly with
the tidyverse to bring the english text into R.2 Here’s what’s happening. First, I obtain the perseus digital library catalog with get_perseus_catalog.
You need to know each text’s Uniform Reference Number (URN) before getting the text. Second, I filter the catalog for New Testament works and grep
for “Gospel”. Third, I iterate through the vector of urns, requesting the English text of each gospel. Fourth, I create the chapter column and return
a data frame. And fifth, I join the perseus catalog back onto our gospels data frame with cleaner labels.
library(rperseus)
library(tidyverse)
library(tidytext)
perseus_catalog <- get_perseus_catalog()
gospels <- perseus_catalog %>%
filter(groupname == "New Testament",
grepl("Gospel", label)) %>%
.$urn %>%
map(get_perseus_text, "eng") %>%
map_df(mutate, chapter = row_number())
gospels <- left_join(gospels, perseus_catalog %>%
mutate(book = stringr::str_replace(label, "Gospel according to ", "")) %>%
select(urn, book), by = "urn")Let’s glance at gospels:
glimpse(gospels)
Observations: 89
Variables: 4
$ urn <chr> "urn:cts:greekLit:tlg0031.tlg003", "urn:cts:greekLit:tlg0031.tlg003", "urn:cts:greek...
$ text <chr> "Since many have undertaken to set in order a narrative concerning those matters whi...
$ chapter <int> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 2...
$ book <chr> "Luke", "Luke", "Luke", "Luke", "Luke", "Luke", "Luke", "Luke", "Luke", "Luke", "Luk...
Those 89 observations correspond to the 89 chapters combined in Matthew, Mark, Luke, and John.
Tidying the Text
With the text in hand, we can unleash tidytext. Much of what follows is adapted from the tidytext vignette, which is brilliant. I added
a few tweaks of my own, though.
The first step is to unnest the word tokens and remove the stop words:
gospel_words <- gospels %>%
unnest_tokens(word, text)
data("stop_words")
gospel_words <- gospel_words %>%
anti_join(stop_words)Sentiment Analysis
Now the fun begins. What are the most frequently appearing joyous words in each gospel? I filtered out “god” so we could see some of the more distinctive words.
gospel_words %>%
filter(word != "god") %>%
semi_join(nrc %>%
filter(sentiment == "joy")) %>%
count(book, word, sort = TRUE) %>%
group_by(book) %>%
slice(1:3)
Source: local data frame [12 x 3]
Groups: book [4]
book word n
<chr> <chr> <int>
1 John love 20
2 John glory 17
3 John true 16
4 Luke found 30
5 Luke blessed 27
6 Luke mother 21
7 Mark mother 19
8 Mark child 11
9 Mark found 10
10 Matthew mother 28
11 Matthew blessed 17
12 Matthew tree 15
Gotta admit, I did not see “tree” coming. Odd. Given the nativity accounts in Matthew and Luke, seeing “mother” is unsurprising, but I didn’t expect to see it in Mark too. “Love”, “glory”, and “true” from John is also unsurprising.
What are the most disgust-inducing words?
Source: local data frame [12 x 3]
Groups: book [4]
book word n
<chr> <chr> <int>
1 John sin 16
2 John flesh 13
3 John death 12
4 Luke woe 15
5 Luke evil 14
6 Luke tree 11
7 Mark unclean 11
8 Mark death 9
9 Mark sick 9
10 Matthew evil 26
11 Matthew tree 15
12 Matthew woe 14
Tree again?! I don’t have an explanation.
Finally, let’s locate the most negative chapter in each gospel:
bingnegative <- get_sentiments("bing") %>%
filter(sentiment == "negative")
wordcounts <- gospel_words %>%
group_by(book, chapter) %>%
summarize(words = n())
gospel_words %>%
semi_join(bingnegative) %>%
group_by(book, chapter) %>%
summarize(negativewords = n()) %>%
left_join(wordcounts, by = c("book", "chapter")) %>%
mutate(ratio = negativewords/words) %>%
filter(chapter != 0) %>%
top_n(1)
Source: local data frame [4 x 5]
Groups: book [4]
book chapter negativewords words ratio
<chr> <int> <int> <int> <dbl>
1 John 9 20 170 0.1176471
2 Luke 6 45 330 0.1363636
3 Mark 5 30 236 0.1271186
4 Matthew 23 37 257 0.1439689
What’s happening in each chapter?
John 9: Jesus heals a man born blind who is then questioned by the Pharisees. The words “sin” and “blind” appear almost every other sentence.
Luke 6: Jesus preaches the devastating Sermon on the Plains, a rebuke to those who are rich, laughing, and full. “Woe” and “judge” appear often.
Mark 5: Three longish stories (for Mark) appear here. First, Jesus confronts the Geresene demonianc and drives out the unclean spirits. Second, Jesus heals the woman with hemorrages. Third, Jesus raises Jairus’ daughter. “Death”, “howling”, and “wailing” define this pericope.
Matthew 23: This was the least surprising. Here Jesus excoriates the Pharisees. I mean, it is brutal. Lots of “woes” and “hypocrites”.
I’d say it checks out! In part II we’ll trot out some visualizations. Stay tuned.