Comparative Text Analysis in R: the LatinX Term according to the Printed Press
I analyzed news articles mentioning the term latinx collected from 6 newspapers in the United States.
Our language is a living construct in which the use -and misuse- of words is a contested field reshaped more than ever by internet culture and the use of digital technologies. Shifts in political and social landscapes in relation to representation, inclusion and identity also play a role in the appearance of new terms, words or phrases. Terms that have arisen relatively recent but are still not fully incorporated into mainstream language are known as neologisms (from the Greek neo = new and logos = speech) and they are usually formed by the combination of existing words, or by the addition to words of novel and unique suffixes or prefixes. The term latinx is such a neologism and is employed to describe people of Latin American descend or cultural identity in the United States. The -x in latinx is intended as a gender-neutral suffix replacing the -o/-a ending of latino and latina ought to Spanish being a grammatically gendered language. The term also accommodates individuals identifying as non-binary and gender non-conforming within the Latin American and Hispanic community.
While the term is gaining recognition, especially in academic settings, its usage is still contested and has become a controversial topic in relation to political representation. The most recent -and perhaps extreme- political controversy arising from the use of the latinx term is the ban of the word from state-government documents in Arkansas and Connecticut. While conservative lawmakers argue the term is an intellectual construct from liberals; some liberal lawmakers argue instead the term is an Americanized word disrespecting the Spanish language and consider the word an act of cultural appropriation.
Because of the above, I performed a comparative text analysis on the latinx term appearing on news articles written by the printed press in United States. I reasoned newspapers tailor their writings to their latinx communities, and these communities are geographically distinct depending on the country-of-origin’s majority. For instance, the composition of the latinx population living in Los Angeles is different from the composition of the latinx population living in New York City and their respective newspapers (Los Angeles Times and The New York Times) would reflect that. Furthermore, newspapers addressing a particular latinx community with Republican and/or Conservative values (think the Miami Herald and Cubans) would have a different take on the latinx term compared to newspapers addressing a latinx community with liberal values (think The New York Times and Puerto Ricans). With this in mind I collected news articles mentioning the term latinx from 6 newspapers and analyzed their text using R. The objective was ‘to extract’ the contextual meaning of the latinx term from each newspaper in order to understand their similitude and differences. The newspapers I included in my study were:
The New York Times > 103 news articles from September-2017 to March-2022 from the 484 news articles available in total until April-22–2022 (21.3% coverage)
The Washington Post > 20 news articles from March-2017 to December-2021 from the 20 news articles available in total until May-17–2022 (100% coverage)
Miami Herald > 78 news articles from July-2016 to March-2022 from the 123 news articles available in total until May-13-2022 (63.4% coverage)
Houston Chronicle > 101 news articles from May-2016 to April-2022 from the 150 news articles available in total until May-4–2022 (67.3% coverage)
Los Angeles Times > 101 news articles from September-2018 to April-2022 from the 866 news articles available in total until May-6–2022 (11.6% coverage)
Chicago Tribune > 31 news articles from September-2020 to May-2022 ( I did not register the total number of news articles that were available until May-2022)
Overall, I collected 434 news articles mentioning the term latinx that were written over a six years time period (July-2016 to May-2022). I collected the news articles ‘manually’ by visiting (and registering as paid subscriber) the webpage of each newspaper, searching for the term latinx and copying the text of each article into a plain text file for each newspaper. The resulting six text files constituted the corpus in my text analysis.
LATIN AMERICAN COUNTRIES THAT WERE MENTIONED DIFFERED FOR EACH NEWSPAPER
The first thing I wanted to test was the relative frequency at which Latin American countries were mentioned by each newspaper. My results showed that the most mentioned countries were Cuba (The Washington Post and Miami Herald), Puerto Rico (The New York Times and Chicago Tribune), and Mexico (Los Angeles Times and Houston Chronicle) respectively (Figure 1). This may suggest that Latinx is a term that may be associated with different communities from different Latin American countries depending on the newspaper, and these communities are mainly derived from three countries only.
# load libraries
library(tidyverse)
library(tidytext)
library(patchwork)
# Visualize mentions of Latin American countries per newspaper
# create a function that tokenizes text
tokenizeText <- function(textFile) {
textFile <- file.choose()
textFile.scanned <- scan(textFile, what = "character", sep = "\n")
textFile.scanned <- paste(textFile.scanned, collapse = " ")
textFile.scanned <- tolower(textFile.scanned)
textFile.scanned.l <- strsplit(textFile.scanned, "[^A-Za-z0-9'_]")
textFile.scanned.v <- unlist(textFile.scanned.l)
notBlanks <- which(textFile.scanned.v != "")
textFile.scanned.v.withoutBlanks <- textFile.scanned.v[notBlanks]
return(textFile.scanned.v.withoutBlanks)
}
# tokenise text from news articles for each newspaper
# I don't know the reason but my tokenizer does not take contractions on the text files I created
# "museum's" is processed as "museum" + "s"
# Washington Post
wp <- tokenizeText(wpText)
wp
# New York Times
nyt <- tokenizeText(nytText)
nyt
# LA Times
lat <- tokenizeText(latText)
lat
# Miami Herald
mihe <- tokenizeText(miheText)
mihe
# Chicago Tribune
chit <- tokenizeText(chitText)
chit
# Houston Chronicle
huc <- tokenizeText(hucText)
huc
# create vectors containing tokens associated with countries in Latin America
Argentina <- c("argentina", "argentineans", "argentines", "argentine", "argentinean")
Bolivia <- c("bolivia", "bolivians", "bolivian")
Chile <- c("chile", "chileans", "chilean")
Colombia <- c("colombia", "colombians", "colombian")
Costa_Rica <- c("costa_rica", "costa_ricans", "costa_rican")
Cuba <- c("cuba", "cubans", "cuban")
Dominican_Republic <- c("dominican", "dominican_republic", "dominicans")
Ecuador <- c("ecuador", "ecuadorians", "ecuadoran")
El_Salvador <- c("el_salvador", "salvadorans", "el_salvadorans", "el_salvadoran", "salvadoran")
Guatemala <- c("guatemala", "guatemalans", "guatemalan")
Honduras <- c("honduras", "hondurans", "honduran")
Mexico <- c("mexico", "mexicans", "chicano", "chicana", "chicanos", "chicanas", "chicanxs", "mexican")
Nicaragua <- c("nicaragua", "nicaraguans", "nicaraguan")
Panama <- c("panama", "panamanians", "panamenian")
Paraguay <- c("paraguay", "paraguayans", "paraguayan")
Peru <- c("peru", "peruvians")
Puerto_Rico <- c("puerto_rico", "puerto_rican", "puerto_ricans", "boricua")
Uruguay <- c("uruguay", "uruguayans", "uruguayan")
Venezuela <- c("venezuela", "venezuelans", "venezuelan")
# calculate relative frequency of mentions for each country
# Washington Post newspaper
argentina.wp <- 100*(length(which(wp == Argentina)) / length(wp))
bolivia.wp <- 100*(length(which(wp == Bolivia)) / length(wp))
chile.wp <- 100*(length(which(wp == Chile)) / length(wp))
colombia.wp <- 100*(length(which(wp == Colombia)) / length(wp))
costarica.wp <- 100*(length(which(wp == Costa_Rica)) / length(wp))
cuba.wp <- 100*(length(which(wp == Cuba)) / length(wp))
dominican.wp <- 100*(length(which(wp == Dominican_Republic)) / length(wp))
ecuador.wp <- 100*(length(which(wp == Ecuador)) / length(wp))
salvador.wp <- 100*(length(which(wp == El_Salvador)) / length(wp))
guatemala.wp <- 100*(length(which(wp == Guatemala)) / length(wp))
honduras.wp <- 100*(length(which(wp == Honduras)) / length(wp))
mexico.wp <- 100*(length(which(wp == Mexico)) / length(wp))
nicaragua.wp <- 100*(length(which(wp == Nicaragua)) / length(wp))
panama.wp <- 100*(length(which(wp == Panama)) / length(wp))
paraguay.wp <- 100*(length(which(wp == Paraguay)) / length(wp))
peru.wp <- 100*(length(which(wp == Peru)) / length(wp))
puertorico.wp <- 100*(length(which(wp == Puerto_Rico)) / length(wp))
uruguay.wp <- 100*(length(which(wp == Uruguay)) / length(wp))
venezuela.wp <- 100*(length(which(wp == Venezuela)) / length(wp))
# create a function to calculate relative frequency of mentions (what I did on the above code)
# country relative frequency as 'corf'
corf <- function(Country, newspaper) {
Country <- 100*(length(which(newspaper == Country)) / length(newspaper))
return(Country)
}
# relative frequency of mentions for each country
# The New York Times
argentina.nyt <- corf(Argentina, nyt)
bolivia.nyt <- corf(Bolivia, nyt)
chile.nyt <- corf(Chile, nyt)
colombia.nyt <- corf(Colombia, nyt)
costarica.nyt <- corf(Costa_Rica, nyt)
cuba.nyt <- corf(Cuba, nyt)
dominican.nyt <- corf(Dominican_Republic, nyt)
ecuador.nyt <- corf(Ecuador, nyt)
salvador.nyt <- corf(El_Salvador, nyt)
guatemala.nyt <- corf(Guatemala, nyt)
honduras.nyt <- corf(Honduras, nyt)
mexico.nyt <- corf(Mexico, nyt)
nicaragua.nyt <- corf(Nicaragua, nyt)
panama.nyt <- corf(Panama, nyt)
paraguay.nyt <- corf(Paraguay, nyt)
peru.nyt <- corf(Peru, nyt)
puertorico.nyt <- corf(Puerto_Rico, nyt)
uruguay.nyt <- corf(Uruguay, nyt)
venezuela.nyt <- corf(Venezuela, nyt)
# LA Times
argentina.lat <- corf(Argentina, lat)
bolivia.lat <- corf(Bolivia, lat)
chile.lat <- corf(Chile, lat)
colombia.lat <- corf(Colombia, lat)
costarica.lat <- corf(Costa_Rica, lat)
cuba.lat <- corf(Cuba, lat)
dominican.lat <- corf(Dominican_Republic, lat)
ecuador.lat <- corf(Ecuador, lat)
salvador.lat <- corf(El_Salvador, lat)
guatemala.lat <- corf(Guatemala, lat)
honduras.lat <- corf(Honduras, lat)
mexico.lat <- corf(Mexico, lat)
nicaragua.lat <- corf(Nicaragua, lat)
panama.lat <- corf(Panama, lat)
paraguay.lat <- corf(Paraguay, lat)
peru.lat <- corf(Peru, lat)
puertorico.lat <- corf(Puerto_Rico, lat)
uruguay.lat <- corf(Uruguay, lat)
venezuela.lat <- corf(Venezuela, lat)
# Miami Herald
argentina.mihe <- corf(Argentina, mihe)
bolivia.mihe <- corf(Bolivia, mihe)
chile.mihe <- corf(Chile, mihe)
colombia.mihe <- corf(Colombia, mihe)
costarica.mihe <- corf(Costa_Rica, mihe)
cuba.mihe <- corf(Cuba, mihe)
dominican.mihe <- corf(Dominican_Republic, mihe)
ecuador.mihe <- corf(Ecuador, mihe)
salvador.mihe <- corf(El_Salvador, mihe)
guatemala.mihe <- corf(Guatemala, mihe)
honduras.mihe <- corf(Honduras, mihe)
mexico.mihe <- corf(Mexico, mihe)
nicaragua.mihe <- corf(Nicaragua, mihe)
panama.mihe <- corf(Panama, mihe)
paraguay.mihe <- corf(Paraguay, mihe)
peru.mihe <- corf(Peru, mihe)
puertorico.mihe <- corf(Puerto_Rico, mihe)
uruguay.mihe <- corf(Uruguay, mihe)
venezuela.mihe <- corf(Venezuela, mihe)
# Chicago Tribune
argentina.chit <- corf(Argentina, chit)
bolivia.chit <- corf(Bolivia, chit)
chile.chit <- corf(Chile, chit)
colombia.chit <- corf(Colombia, chit)
costarica.chit <- corf(Costa_Rica, chit)
cuba.chit <- corf(Cuba, chit)
dominican.chit <- corf(Dominican_Republic, chit)
ecuador.chit <- corf(Ecuador, chit)
salvador.chit <- corf(El_Salvador, chit)
guatemala.chit <- corf(Guatemala, chit)
honduras.chit <- corf(Honduras, chit)
mexico.chit <- corf(Mexico, chit)
nicaragua.chit <- corf(Nicaragua, chit)
panama.chit <- corf(Panama, chit)
paraguay.chit <- corf(Paraguay, chit)
peru.chit <- corf(Peru, chit)
puertorico.chit <- corf(Puerto_Rico, chit)
uruguay.chit <- corf(Uruguay, chit)
venezuela.chit <- corf(Venezuela, chit)
# Huston Chronicle
argentina.huc <- corf(Argentina, huc)
bolivia.huc <- corf(Bolivia, huc)
chile.huc <- corf(Chile, huc)
colombia.huc <- corf(Colombia, huc)
costarica.huc <- corf(Costa_Rica, huc)
cuba.huc <- corf(Cuba, huc)
dominican.huc <- corf(Dominican_Republic, huc)
ecuador.huc <- corf(Ecuador, huc)
salvador.huc <- corf(El_Salvador, huc)
guatemala.huc <- corf(Guatemala, huc)
honduras.huc <- corf(Honduras, huc)
mexico.huc <- corf(Mexico, huc)
nicaragua.huc <- corf(Nicaragua, huc)
panama.huc <- corf(Panama, huc)
paraguay.huc <- corf(Paraguay, huc)
peru.huc <- corf(Peru, huc)
puertorico.huc <- corf(Puerto_Rico, huc)
uruguay.huc <- corf(Uruguay, huc)
venezuela.huc <- corf(Venezuela, huc)
# start constructing data frame
LatinAmericanCountries <- c("Argentina", "Bolivia", "Chile", "Colombia", "Costa_Rica", "Cuba",
"Dominican_Republic", "Ecuador", "El_Salvador", "Guatemala", "Honduras",
"Mexico", "Nicaragua", "Panama", "Paraguay", "Peru", "Puerto_Rico",
"Uruguay", "Venezuela")
Washington_Post <- c(argentina.wp, bolivia.wp, chile.wp, colombia.wp, costarica.wp, cuba.wp,
dominican.wp, ecuador.wp, salvador.wp, guatemala.wp, honduras.wp, mexico.wp,
nicaragua.wp, panama.wp, paraguay.wp, peru.wp, puertorico.wp, uruguay.wp, venezuela.wp)
New_York_Times <- c(argentina.nyt, bolivia.nyt, chile.nyt, colombia.nyt, costarica.nyt, cuba.nyt, dominican.nyt,
ecuador.nyt, salvador.nyt, guatemala.nyt, honduras.nyt, mexico.nyt, nicaragua.nyt,
panama.nyt, paraguay.nyt, peru.nyt, puertorico.nyt, uruguay.nyt, venezuela.nyt)
Los_Angeles_Times <- c(argentina.lat, bolivia.lat, chile.lat, colombia.lat, costarica.lat, cuba.lat,
dominican.lat, ecuador.lat, salvador.lat, guatemala.lat, honduras.lat,
mexico.lat, nicaragua.lat, panama.lat, paraguay.lat, peru.lat, puertorico.lat,
uruguay.lat, venezuela.lat)
Miami_Herald <- c(argentina.mihe, bolivia.mihe, chile.mihe, colombia.mihe, costarica.mihe, cuba.mihe,
dominican.mihe, ecuador.mihe, salvador.mihe, guatemala.mihe, honduras.mihe, mexico.mihe,
nicaragua.mihe, panama.mihe, paraguay.mihe, peru.mihe, puertorico.mihe, uruguay.mihe, venezuela.mihe)
Chicago_Tribune <- c(argentina.chit, bolivia.chit, chile.chit, colombia.chit, costarica.chit, cuba.chit,
dominican.chit, ecuador.chit, salvador.chit, guatemala.chit, honduras.chit, mexico.chit,
nicaragua.chit, panama.chit, paraguay.chit, peru.chit, puertorico.chit, uruguay.chit,
venezuela.chit)
Houston_Chronicle <- c(argentina.huc, bolivia.huc, chile.huc, colombia.huc, costarica.huc, cuba.huc,
dominican.huc, ecuador.huc, salvador.huc, guatemala.huc, honduras.huc, mexico.huc,
nicaragua.huc, panama.huc, paraguay.huc, peru.huc, puertorico.huc, uruguay.huc,
venezuela.huc)
# data frame Countries relative frequency per newspaper as 'corelfren'
corelfren <- data.frame(LatinAmericanCountries, Washington_Post, New_York_Times, Los_Angeles_Times,
Miami_Herald, Chicago_Tribune, Houston_Chronicle)
View(corelfren)
# Visuallize results
# Determine the overall limits for x
x_limits <- range(c(0, 0.07))
# plot relative frequencies for the Washington Post
wp <- ggplot(corelfren, aes(x = Washington_Post, y = LatinAmericanCountries)) +
geom_col() +
scale_x_continuous(breaks = seq(0, 0.07, 0.01)) +
theme_minimal()
wp <- wp + coord_cartesian(xlim = x_limits)
# plot relative frequencies for the New York Times
nyt <- ggplot(corelfren, aes(x = New_York_Times, y = LatinAmericanCountries)) +
geom_col() +
scale_x_continuous(breaks = seq(0, 0.07, 0.01)) +
theme_minimal()
nyt <- nyt + coord_cartesian(xlim = x_limits)
# plot relative frequencies for the Los Angeles Times
lat <- ggplot(corelfren, aes(x = Los_Angeles_Times, y = LatinAmericanCountries)) +
geom_col() +
scale_x_continuous(breaks = seq(0, 0.07, 0.01)) +
theme_minimal()
lat <- lat + coord_cartesian(xlim = x_limits)
# plot relative frequencies for the Miami Herald
mihe <- ggplot(corelfren, aes(x = Miami_Herald, y = LatinAmericanCountries)) +
geom_col() +
scale_x_continuous(breaks = seq(0, 0.07, 0.01)) +
theme_minimal()
mihe <- mihe + coord_cartesian(xlim = x_limits)
# plot relative frequencies for Chicago Tribune
chica <- ggplot(corelfren, aes(x = Chicago_Tribune, y = LatinAmericanCountries)) +
geom_col() +
scale_x_continuous(breaks = seq(0, 0.07, 0.01)) +
theme_minimal()
chica <- chica + coord_cartesian(xlim = x_limits)
# plot relative frequencies for Houston Chronicle
hucro <- ggplot(corelfren, aes(x = Houston_Chronicle, y = LatinAmericanCountries)) +
geom_col() +
scale_x_continuous(breaks = seq(0, 0.07, 0.01)) +
theme_minimal()
hucro <- hucro + coord_cartesian(xlim = x_limits)
# plot results
wp + nyt + lat + mihe + chica + hucro
# save plot as .svg and modify it in Adobe Illustrator for better visualization
ggsave("newspapers_relFreq.svg", width = 4000, height = 3000, units = "px")
WHAT DID THE NEWSPAPERS WRITE ABOUT? COMPARATIVE TOPIC MODELING
In order to compare the topics wrote about by the different newspapers, I implemented topic modeling using the mallet package. Topic modeling is an unsupervised machine learning method used in the field of natural language processing (nlp) and consists of identifying words in a corpus of documents that co-occur together and thus share semantic relationship. I used Latent Dirichlet Allocation (LDA) algorithm that assumes each document is a collection of topics in different proportions. For this I chunked the text from each newspaper into 1,000 words segments and extracted 6 topics (weighted word distributions) (Figure 2A). The first topic seemed to be about health and housing related issues from the afro-latinx community; the second topic seemed to be about politics and political representation of latinxs; the third topic seemed to be about academic life of latinxs in Houston; the fourth topic seemed to be identity of latinx; the fifth topic seemed to be about music, theater and other cultural activities of the latinx community in Miami; and the sixth topic seemed to be about the art scene and cultural representation of the latinx community mainly in New York.
I subsequently examined the relative importance of each topic per newspaper document and found that topic 1 was more relevant in news articles from Chicago Tribune; topic 2 was more relevant in news articles from Miami Herald; topic 3 was more relevant in news articles from The Houston Chronicle; and topic 6 was more relevant in news articles from The New York Times. The topics 4 and 5 were relatively relevant across the six newspapers examined (Figure 2B).
For me it was interesting to find that topic 2, which was about politics and political representation of latinxs was most relevant in news articles from the Miami Herald. It is known that Florida is a politically charged state and Republican oriented in political terms.
# TOPIC MODELING
# load libraries
library(mallet)
library(wordcloud)
library(tidyverse)
library(patchwork)
# place .R file in same folder as .txt files
input_dir <- "/Users/martincalvinotorterolo/Desktop/latinX/data"
files_v <- dir(input_dir, "\\.txt$")
files_v
# create empty data frame
documents_df <- NULL
# set chunk size > I will extract/learn the topics from each text segment of 1,000 tokens
chunk_size <- 1000
# loop over each text file
for (i in seq_along(files_v)) {
text_v <- scan(files_v[i], what = "character", sep = "\n")
text_v <- paste(text_v, collapse = " ")
text_words_v <- tolower(text_v)
text_words_v <- strsplit(text_words_v, "[^A-Za-z0-9_']")
text_words_v <- unlist(text_words_v)
text_words_v <- text_words_v[which(text_words_v != "")]
x <- seq_along(text_words_v)
chunks_l <- split(text_words_v, ceiling(x / chunk_size))
if (length(chunks_l[[length(chunks_l)]]) <= chunk_size / 2) {
chunks_l[[length(chunks_l) - 1]] <- c(chunks_l[[length(chunks_l) - 1]], chunks_l[[length(chunks_l)]])
chunks_l[[length(chunks_l)]] <- NULL
}
chunk_strings_l <- lapply(chunks_l, paste, collapse = " ")
chunks_df <- do.call(rbind, chunk_strings_l)
textname_v <- gsub("\\..*", "", files_v[i])
chunk_ids_v <- 1:nrow(chunks_df)
chunk_names_v <- paste(textname_v, chunk_ids_v, sep = "_")
file_df <- data.frame(id = chunk_names_v, text = chunks_df, stringsAsFactors = FALSE)
documents_df <- rbind(documents_df, file_df)
}
# inspect data frame
head(documents_df)
View(documents_df)
# "id" is the number for each chunk per newspaper text, and "text" contains the 1,000 tokens text segment
colnames(documents_df)
# instantiate a mallet object
mallet.instances <- mallet.import(documents_df$id, documents_df$text,
"/Users/martincalvinotorterolo/Desktop/latinX/data/stopwords.csv",
FALSE, token.regexp = "[\\p{L}']+")
# create a topic model trainer object
# set the number of topics the model will contain to 6
topic_model <- MalletLDA(num.topics = 6)
# load text data for modeling
topic_model$loadDocuments(mallet.instances)
# access the list of the vocabulary after stop words were removed
vocabulary <- topic_model$getVocabulary()
vocabulary[1:200]
# access info about the frequency of words in the corpus
word_freqs <- mallet.word.freqs(topic_model)
head(word_freqs, n = 50)
### train the topic model with 1,000 iterations
topic_model$train(1000)
# unpack the model
# extract words per topic
# create a matrix of 6 rows (one for each topic extracted) and each column a unique word type in the corpus
topic_words_m <- mallet.topic.words(topic_model, smoothed = TRUE, normalized = TRUE)
topic_words_m[1:6, 1:10]
dim(topic_words_m) # a matrix of 6 rows x 21,511 columns
colnames(topic_words_m) <- vocabulary
topic_words_m[1:6, 1:10]
rowSums(topic_words_m) # because normalized = TRUE was used as argument to mallet.topic.words() function
# the values in each topic row were converted to percentages that sum to 1
# compare the relative weight of specific word types as a percentage of each topic
keywords <- c("pandemic", "covid", "virus", "health")
topic_words_m[, keywords]
# now examine the 100 most heavily weighted words in each topic
topic1_top_words <- mallet.top.words(topic_model, topic_words_m[1, ], 200)
topic1_top_words # about the pandemic and health
topic2_top_words <- mallet.top.words(topic_model, topic_words_m[2, ], 200)
topic2_top_words # about political representation
topic3_top_words <- mallet.top.words(topic_model, topic_words_m[3, ], 200)
topic3_top_words # about student life, university activities by latinos
topic4_top_words <- mallet.top.words(topic_model, topic_words_m[4, ], 200)
topic4_top_words # social justice and latinos
topic5_top_words <- mallet.top.words(topic_model, topic_words_m[5, ], 200)
topic5_top_words # theater. music
topic6_top_words <- mallet.top.words(topic_model, topic_words_m[6, ], 200)
topic6_top_words # art, books and films
# topic visualization using word clouds
# Topic1
set.seed(1234)
wordcloud(words = topic1_top_words$term, freq = topic1_top_words$weight, min.freq = 1,
max.words=200, random.order=FALSE, rot.per=0.35, scale=c(3.5, 0))
# Topic 2
set.seed(1234)
wordcloud(words = topic2_top_words$term, freq = topic2_top_words$weight, min.freq = 1,
max.words=200, random.order=FALSE, rot.per=0.35, scale=c(3.4, 0.1))
# Topic 3
set.seed(1234)
wordcloud(words = topic3_top_words$term, freq = topic3_top_words$weight, min.freq = 1,
max.words=200, random.order=FALSE, rot.per=0.35, scale=c(3.5, 0.1))
# Topic 4
set.seed(1234)
wordcloud(words = topic4_top_words$term, freq = topic4_top_words$weight, min.freq = 1,
max.words=200, random.order=FALSE, rot.per=0.35, scale=c(3.6, 0.4))
# Topic 5
set.seed(1234)
wordcloud(words = topic5_top_words$term, freq = topic5_top_words$weight, min.freq = 1,
max.words=200, random.order=FALSE, rot.per=0.35, scale=c(3.7, 0.35))
# Topic 6
set.seed(1234)
wordcloud(words = topic6_top_words$term, freq = topic6_top_words$weight, min.freq = 1,
max.words=200, random.order=FALSE, rot.per=0.35, scale=c(3.5, 0.35))
# inspect the probability of each topic appearing on each document (composed of news from each newspaper)
# assess the proportion of a document that is about each topic
doc_topic_m <- mallet.doc.topics(topic_model, smoothed = TRUE, normalized = TRUE)
doc_topic_m # each column is a topic and each row is a text chunk 1,000 tokens long
# look at the document from each newspaper as a whole
# calculate mean topical values
file_ids_v <- documents_df[, 1]
head(file_ids_v)
# modify documents_df as to be able to use group_by() function later on
file_id_l <- strsplit(file_ids_v, "_")
file_chunk_id_l <- lapply(file_id_l, rbind)
file_chunk_id_m <- do.call(rbind, file_chunk_id_l)
head(file_chunk_id_m)
# convert matrix to data frame
doc_topics_df <- as.data.frame(doc_topic_m)
doc_topics_df <- data.frame(file = file_chunk_id_m[,1],
doc_topics_df,
stringsAsFactors = FALSE)
# calculate the proportion of topics per newspaper document
doc_topic_means_df <- group_by(doc_topics_df, file) %>%
summarize_all(mean)
View(doc_topic_means_df)
# visualize results
# set the same scale for the x-axis
x_limits <- range(c(0, 0.60))
# topic 1
t1 <- ggplot(data = doc_topic_means_df, mapping = aes(x = V1, y = file)) +
geom_col() +
scale_x_continuous(breaks = seq(0, 0.60, 0.10)) +
coord_flip() +
theme_minimal()
t1 <- t1 + coord_cartesian(xlim = x_limits)
t1
# topic 2
t2 <- ggplot(data = doc_topic_means_df, mapping = aes(x = V2, y = file)) +
geom_col() +
scale_x_continuous(breaks = seq(0, 0.60, 0.10)) +
coord_flip() +
theme_minimal()
t2 <- t2 + coord_cartesian(xlim = x_limits)
t2
# topic 3
t3 <- ggplot(data = doc_topic_means_df, mapping = aes(x = V3, y = file)) +
geom_col() +
scale_x_continuous(breaks = seq(0, 0.60, 0.10)) +
coord_flip() +
theme_minimal()
t3 <- t3 + coord_cartesian(xlim = x_limits)
t3
# topic 4
t4 <- ggplot(data = doc_topic_means_df, mapping = aes(x = V4, y = file)) +
geom_col() +
scale_x_continuous(breaks = seq(0, 0.60, 0.10)) +
coord_flip() +
theme_minimal()
t4 <- t4 + coord_cartesian(xlim = x_limits)
t4
# topic 5
t5 <- ggplot(data = doc_topic_means_df, mapping = aes(x = V5, y = file)) +
geom_col() +
scale_x_continuous(breaks = seq(0, 0.60, 0.10)) +
coord_flip() +
theme_minimal()
t5 <- t5 + coord_cartesian(xlim = x_limits)
t5
# topic 6
t6 <- ggplot(data = doc_topic_means_df, mapping = aes(x = V6, y = file)) +
geom_col() +
scale_x_continuous(breaks = seq(0, 0.60, 0.10)) +
coord_flip() +
theme_minimal()
t6 <- t6 + coord_cartesian(xlim = x_limits)
t6
# create composite plot
t1 + t2 + t3 + t4 + t5 + t6
# save plot, export it as .svg and modify it accordingly in Adobe Illustrator
ggsave("topics_per_newspaper.svg", width = 4000, height = 3000, units = "px")
KEYWORD-IN-CONTEXT ANALYSIS: EXTRACTING THE MEANING OF THE TERM LATINX
You can understand a word’s meaning, or sense, by inspecting the other words around it. Keyword-in-context searches are computational strategies to extract a certain amount of words written before and after the appearance of a keyword throughout the text. In my work I extracted the word context associated with the term latinx for each newspaper document by inspecting five words before and five words after each time latinx was mentioned (Figure 3). This allowed me to compare the words that were most frequently associated to the latinx term and how they differed among newspapers (Figure 4).
# KEYWORD-IN-CONTEXT
# show path to newspaper documents
input_dir <- "/Users/martincalvinotorterolo/Desktop/latinX/data"
files_v <- dir(input_dir, "\\.txt$", full.names = TRUE)
files_v
# create function to show files' paths
show_files <- function(directory_path, pattern = "\\.txt$") {
# function to print a vector of file paths and their index numbers in user-friendly format
file_name_v <- dir(directory_path, pattern, full.names = TRUE)
for(i in seq_along(file_name_v)) {
cat(i, file_name_v[i], "\n", sep = " ")
}
}
# show files's paths
show_files(input_dir)
# create function to tokenize text in newspaper document
make_token_v <- function(file_path, pattern = "\\W") {
# function returns an ordered vector of tokens from the file referenced in the file_path argument
text_v <- scan(file_path, what = "character", sep = "\n")
text_v <- paste(text_v, collapse = " ")
text_lower_v <- tolower(text_v)
text_words_v <- strsplit(text_lower_v, pattern)
text_words_v <- unlist(text_words_v)
text_words_v <- text_words_v[which(text_words_v != "")]
return(text_words_v)
}
# NEW YORK TIMES
# tokenize news articles from THe New York Times document
latinx_nyt <- make_token_v("/Users/martincalvinotorterolo/Desktop/latinX/data/NYTimes.txt")
# inspect tokens
latinx_nyt
# save positions corresponding to each mention of latinx
positions_latinx_nyt <- which(latinx_nyt == "latinx")
# inspect positions
positions_latinx_nyt
length(positions_latinx_nyt) # 292 mentions of the latinx term
# inspect the latinx's mention number 432 and its word context as an example
latinx_nyt[432]
latinx_nyt[427:437]
# extract the word context for each mention of latinx in NYTimes
for (i in 1:length(positions_latinx_nyt)) {
cat(latinx_nyt[(positions_latinx_nyt[i]-5):(positions_latinx_nyt[i]-1)], "[", latinx_nyt[positions_latinx_nyt[i]], "]",
latinx_nyt[(positions_latinx_nyt[i]+1):(positions_latinx_nyt[i]+5)], "\n", sep = " ")
}
# save the console output of the above code into a text file
# open connection to a text file
sink(file = "NYTimes_kwic_latinx.txt")
# execute code to extract keyword in context
for (i in 1:length(positions_latinx_nyt)) {
cat(latinx_nyt[(positions_latinx_nyt[i]-5):(positions_latinx_nyt[i]-1)], "[", latinx_nyt[positions_latinx_nyt[i]], "]",
latinx_nyt[(positions_latinx_nyt[i]+1):(positions_latinx_nyt[i]+5)], "\n", sep = " ")
}
# close connection to text file
sink()
# LOS ANGELES TIMES
# tokenize news articles from Los Angeles Times document
latinx_lat <- make_token_v("/Users/martincalvinotorterolo/Desktop/latinX/data/LATimes.txt")
# inspect tokens
latinx_lat
# save positions corresponding to each mention of latinx
positions_latinx_lat <- which(latinx_lat == "latinx")
# inspect positions
positions_latinx_lat
length(positions_latinx_lat) # 527 mentions of the latinx term
# inspect the latinx's mention number 729 and its word context as an example
latinx_lat[729]
latinx_lat[724:734]
# extract the word context for each mention of latinx in Los Angeles Times
for (i in 1:length(positions_latinx_lat)) {
cat(latinx_lat[(positions_latinx_lat[i]-5):(positions_latinx_lat[i]-1)], "[", latinx_lat[positions_latinx_lat[i]], "]",
latinx_lat[(positions_latinx_lat[i]+1):(positions_latinx_lat[i]+5)], "\n", sep = " ")
}
# save the console output of the above code into a text file
# open connection to a text file
sink(file = "LATimes_kwic_latinx.txt")
# execute code to extract keyword in context
for (i in 1:length(positions_latinx_lat)) {
cat(latinx_lat[(positions_latinx_lat[i]-5):(positions_latinx_lat[i]-1)], "[", latinx_lat[positions_latinx_lat[i]], "]",
latinx_lat[(positions_latinx_lat[i]+1):(positions_latinx_lat[i]+5)], "\n", sep = " ")
}
# close connection to text file
sink()
# HOUSTON CHRONICLE
# tokenize news articles from Houston Chronicle document
latinx_hc <- make_token_v("/Users/martincalvinotorterolo/Desktop/latinX/data/HoustonChronicle.txt")
# inspect tokens
latinx_hc
# save positions corresponding to each mention of latinx
positions_latinx_hc <- which(latinx_hc == "latinx")
# inspect positions
positions_latinx_hc
length(positions_latinx_hc) # 207 mentions of the latinx term
# inspect the latinx's mention number 9497 and its word context as an example
latinx_hc[9497]
latinx_hc[9492:9502]
# extract the word context for each mention of latinx in Houston Chronicle
for (i in 1:length(positions_latinx_hc)) {
cat(latinx_hc[(positions_latinx_hc[i]-5):(positions_latinx_hc[i]-1)], "[", latinx_hc[positions_latinx_hc[i]], "]",
latinx_hc[(positions_latinx_hc[i]+1):(positions_latinx_hc[i]+5)], "\n", sep = " ")
}
# save the console output of the above code into a text file
# open connection to a text file
sink(file = "HC_kwic_latinx.txt")
# execute code to extract keyword in context
for (i in 1:length(positions_latinx_hc)) {
cat(latinx_hc[(positions_latinx_hc[i]-5):(positions_latinx_hc[i]-1)], "[", latinx_hc[positions_latinx_hc[i]], "]",
latinx_hc[(positions_latinx_hc[i]+1):(positions_latinx_hc[i]+5)], "\n", sep = " ")
}
# close connection to text file
sink()
# THE WASHINGTON POST
# tokenize news articles from The Washington Post document
latinx_wp <- make_token_v("/Users/martincalvinotorterolo/Desktop/latinX/data/WashingtonPost.txt")
# inspect tokens
latinx_wp
# save positions corresponding to each mention of latinx
positions_latinx_wp <- which(latinx_wp == "latinx")
# inspect positions
positions_latinx_wp
length(positions_latinx_wp) # 186 mentions of the latinx term
# inspect the latinx's mention number 160 and its word context as an example
latinx_wp[160]
latinx_wp[155:165]
# extract the word context for each mention of latinx in The Washington Post
for (i in 1:length(positions_latinx_wp)) {
cat(latinx_wp[(positions_latinx_wp[i]-5):(positions_latinx_wp[i]-1)], "[", latinx_wp[positions_latinx_wp[i]], "]",
latinx_wp[(positions_latinx_wp[i]+1):(positions_latinx_wp[i]+5)], "\n", sep = " ")
}
# save the console output of the above code into a text file
# open connection to a text file
sink(file = "WP_kwic_latinx.txt")
# execute code to extract keyword in context
for (i in 1:length(positions_latinx_wp)) {
cat(latinx_wp[(positions_latinx_wp[i]-5):(positions_latinx_wp[i]-1)], "[", latinx_wp[positions_latinx_wp[i]], "]",
latinx_wp[(positions_latinx_wp[i]+1):(positions_latinx_wp[i]+5)], "\n", sep = " ")
}
# close connection to text file
sink()
# MIAMI HERALD
# tokenize news articles from Miami Herald document
latinx_mihe <- make_token_v("/Users/martincalvinotorterolo/Desktop/latinX/data/MiamiHerald.txt")
# inspect tokens
latinx_mihe
# save positions corresponding to each mention of latinx
positions_latinx_mihe <- which(latinx_mihe == "latinx")
# inspect positions
positions_latinx_mihe
length(positions_latinx_mihe) # 133 mentions of the latinx term
# inspect the latinx's mention number 381 and its word context as an example
latinx_mihe[381]
latinx_mihe[376:386]
# extract the word context for each mention of latinx in Miami Herald
for (i in 1:length(positions_latinx_mihe)) {
cat(latinx_mihe[(positions_latinx_mihe[i]-5):(positions_latinx_mihe[i]-1)], "[", latinx_mihe[positions_latinx_mihe[i]], "]",
latinx_mihe[(positions_latinx_mihe[i]+1):(positions_latinx_mihe[i]+5)], "\n", sep = " ")
}
# save the console output of the above code into a text file
# open connection to a text file
sink(file = "MIHE_kwic_latinx.txt")
# execute code to extract keyword in context
for (i in 1:length(positions_latinx_mihe)) {
cat(latinx_mihe[(positions_latinx_mihe[i]-5):(positions_latinx_mihe[i]-1)], "[", latinx_mihe[positions_latinx_mihe[i]], "]",
latinx_mihe[(positions_latinx_mihe[i]+1):(positions_latinx_mihe[i]+5)], "\n", sep = " ")
}
# close connection to text file
sink()
# CHICAGO TRIBUNE
# tokenize news articles from Chicago Tribune document
latinx_chi <- make_token_v("/Users/martincalvinotorterolo/Desktop/latinX/data/ChicagoTribune.txt")
# inspect tokens
latinx_chi
# save positions corresponding to each mention of latinx
positions_latinx_chi <- which(latinx_chi == "latinx")
# inspect positions
positions_latinx_chi
length(positions_latinx_chi) # 126 mentions of the latinx term
# inspect the latinx's mention at position 637 and its word context as an example
latinx_chi[637]
latinx_chi[632:642]
# extract the word context for each mention of latinx in Miami Herald
for (i in 1:length(positions_latinx_chi)) {
cat(latinx_chi[(positions_latinx_chi[i]-5):(positions_latinx_chi[i]-1)], "[", latinx_chi[positions_latinx_chi[i]], "]",
latinx_chi[(positions_latinx_chi[i]+1):(positions_latinx_chi[i]+5)], "\n", sep = " ")
}
# save the console output of the above code into a text file
# open connection to a text file
sink(file = "CHI_kwic_latinx.txt")
# execute code to extract keyword in context
for (i in 1:length(positions_latinx_chi)) {
cat(latinx_chi[(positions_latinx_chi[i]-5):(positions_latinx_chi[i]-1)], "[", latinx_chi[positions_latinx_chi[i]], "]",
latinx_chi[(positions_latinx_chi[i]+1):(positions_latinx_chi[i]+5)], "\n", sep = " ")
}
# close connection to text file
sink()
# MOST FREQUENT WORDS IN-CONTEXT-WITH LATINX
# tokenize keyword-in-context text files created earlier for each newspaper
# use tokenizer function created earlier
tokenizeText <- function(textFile) {
textFile <- file.choose()
textFile.scanned <- scan(textFile, what = "character", sep = "\n")
textFile.scanned <- paste(textFile.scanned, collapse = " ")
textFile.scanned <- tolower(textFile.scanned)
textFile.scanned.l <- strsplit(textFile.scanned, "[^A-Za-z0-9_']")
textFile.scanned.v <- unlist(textFile.scanned.l)
notBlanks <- which(textFile.scanned.v != "")
textFile.scanned.v.withoutBlanks <- textFile.scanned.v[notBlanks]
return(textFile.scanned.v.withoutBlanks)
}
### Los Angeles Times
lat <- tokenizeText(textFile)
lat
length(lat) # 5,803 tokens
# create table with token counts
lat_t <- table(lat)
View(lat_t)
# create sorted table
sor_lat_t <- sort(lat_t, decreasing = TRUE)
View(sor_lat_t)
# calculate relative frequency for each token
rf_lat <- 100*(sor_lat_t / length(lat))
View(rf_lat)
# convert relative frequency table to data frame
df_rf_lat <- as.data.frame(rf_lat)
# remove stop words from df_rf_lat
stop_words <- read.csv(file.choose()) # access stopwords.csv file
colnames(stop_words)
View(stop_words)
# create vector of stop words and add new ones
stop_words_vector <- c(stop_words$words, "latinx", "s", "t", "among", "like", "term", "non", "using", "l", "many")
# filter out stop words
df_rf_lat <- filter(df_rf_lat, !(lat %in% stop_words_vector))
View(df_rf_lat)
# plot results for top 25 most frequent words in context associated with latinx
lat_plot <- ggplot(data = df_rf_lat[1:25, ], aes(x = lat, y = Freq)) +
geom_col() +
coord_flip() +
theme_bw()
# inspect plot
lat_plot
### The New York Times
nyt <- tokenizeText(textFile)
nyt
length(nyt) # 3,217 tokens
# create table with token counts
nyt_t <- table(nyt)
View(nyt_t)
# create sorted table
sor_nyt_t <- sort(nyt_t, decreasing = TRUE)
View(sor_nyt_t)
# calculate relative frequency for each token
rf_nyt <- 100*(sor_nyt_t / length(nyt))
View(rf_nyt)
# convert relative frequency table to data frame
df_rf_nyt <- as.data.frame(rf_nyt)
# create vector of stop words and add new ones
stop_words_vector <- c(stop_words$words, "latinx", "s", "t", "among", "like", "term", "non", "using", "l", "many")
# filter out stop words
df_rf_nyt <- filter(df_rf_nyt, !(nyt %in% stop_words_vector))
View(df_rf_nyt)
# plot results for top 25 most frequent words in context associated with latinx
nyt_plot <- ggplot(data = df_rf_nyt[1:25, ], aes(x = nyt, y = Freq)) +
geom_col() +
coord_flip() +
theme_bw()
# inspect plot
nyt_plot
### Houston Chronicle
huc <- tokenizeText(textFile)
huc
length(huc) # 2,279 tokens
# create table with token counts
huc_t <- table(huc)
View(huc_t)
# create sorted table
sor_huc_t <- sort(huc_t, decreasing = TRUE)
View(sor_huc_t)
# calculate relative frequency for each token
rf_huc <- 100*(sor_huc_t / length(huc))
View(rf_huc)
# convert relative frequency table to data frame
df_rf_huc <- as.data.frame(rf_huc)
# create vector of stop words and add new ones
stop_words_vector <- c(stop_words$words, "latinx", "s", "t", "among", "like", "term", "non", "using", "l", "many",
"percent", "times")
# filter out stop words
df_rf_huc <- filter(df_rf_huc, !(huc %in% stop_words_vector))
View(df_rf_huc)
# plot results for top 25 most frequent words in context associated with latinx
huc_plot <- ggplot(data = df_rf_huc[1:25, ], aes(x = huc, y = Freq)) +
geom_col() +
coord_flip() +
theme_bw()
# inspect plot
huc_plot
### The Washington Post
wap <- tokenizeText(textFile)
wap
length(wap) # 2,046 tokens
# create table with token counts
wap_t <- table(wap)
View(wap_t)
# create sorted table
sor_wap_t <- sort(wap_t, decreasing = TRUE)
View(sor_wap_t)
# calculate relative frequency for each token
rf_wap <- 100*(sor_wap_t / length(wap))
View(rf_wap)
# convert relative frequency table to data frame
df_rf_wap <- as.data.frame(rf_wap)
# create vector of stop words and add new ones
stop_words_vector <- c(stop_words$words, "latinx", "s", "t", "among", "like", "term", "non", "using", "l", "many",
"percent", "times", "used", "heard", "despite", "often")
# filter out stop words
df_rf_wap <- filter(df_rf_wap, !(wap %in% stop_words_vector))
View(df_rf_wap)
# plot results for top 25 most frequent words in context associated with latinx
wap_plot <- ggplot(data = df_rf_wap[1:25, ], aes(x = wap, y = Freq)) +
geom_col() +
coord_flip() +
theme_bw()
# inspect plot
wap_plot
### Miami Herald
mih <- tokenizeText(textFile)
mih
length(mih) # 1,463 tokens
# create table with token counts
mih_t <- table(mih)
View(mih_t)
# create sorted table
sor_mih_t <- sort(mih_t, decreasing = TRUE)
View(sor_mih_t)
# calculate relative frequency for each token
rf_mih <- 100*(sor_mih_t / length(mih))
View(rf_mih)
# convert relative frequency table to data frame
df_rf_mih <- as.data.frame(rf_mih)
# create vector of stop words and add new ones
stop_words_vector <- c(stop_words$words, "latinx", "s", "t", "among", "like", "term", "non", "using", "l", "many",
"percent", "times", "used", "heard", "despite", "often", "called", "4")
# filter out stop words
df_rf_mih <- filter(df_rf_mih, !(mih %in% stop_words_vector))
View(df_rf_mih)
# plot results for top 25 most frequent words in context associated with latinx
mih_plot <- ggplot(data = df_rf_mih[1:25, ], aes(x = mih, y = Freq)) +
geom_col() +
coord_flip() +
theme_bw()
# inspect plot
mih_plot
### Chicago Tribune
chit <- tokenizeText(textFile)
chit
length(chit) # 1,387 tokens
# create table with token counts
chit_t <- table(chit)
View(chit_t)
# create sorted table
sor_chit_t <- sort(chit_t, decreasing = TRUE)
View(sor_chit_t)
# calculate relative frequency for each token
rf_chit <- 100*(sor_chit_t / length(chit))
View(rf_chit)
# convert relative frequency table to data frame
df_rf_chit <- as.data.frame(rf_chit)
# create vector of stop words and add new ones
stop_words_vector <- c(stop_words$words, "latinx", "s", "t", "among", "like", "term", "non", "using", "l", "many",
"percent", "times", "used", "heard", "despite", "often", "called", "4", "re")
# filter out stop words
df_rf_chit <- filter(df_rf_chit, !(chit %in% stop_words_vector))
View(df_rf_chit)
# plot results for top 25 most frequent words in context associated with latinx
chit_plot <- ggplot(data = df_rf_chit[1:25, ], aes(x = chit, y = Freq)) +
geom_col() +
coord_flip() +
theme_bw()
# inspect plot
chit_plot
### create composite plot
lat_plot + nyt_plot + huc_plot + wap_plot + mih_plot + chit_plot
# save plot as .svg and modify it in Adobe Illustrator for better visualization
ggsave("mostFrequent_words_in_context.svg", width = 4000, height = 3000, units = "px")
I found that words such as black, community (communities) and people were among the most frequently associated with the latinx term. This is because latinx was mentioned as minority group in the expression “black and latinx community/communities” or referred as “latinx people”. These words were present across the six newspapers I studied.
In news articles from Los Angeles Times, the latinx term was primarily associated with words related to the moving image industry (stories, writers, directors, tv, creators, hollywood) and the representation of latinx within it. In news articles from The New York Times, the latinx term was associated with words related to cultural production in the visual and performing arts (art, writers, artists, performers, culture), and political representation (voters). In news articles from the Houston Chronicle, the latinx term was associated with words related to college life (students, fellowship), identity (ethnic, conflict, chicano, identity, immigrants) and also art (art, artists). In news articles from The Washington Post, words frequently associated with latinx were related to identity (identity, identify, racial, gender, immigrants) and political representation (debate, voters). Similarly, words associated with the latinx term in news articles from the Miami Herald were related to identity (gender, lgbtq, women, african, white) and political representation (voters, organization). Finally, in news articles from the Chicago Tribune, words associated with the latinx term were related to religion (nonreligious, catholic) and this aspect differed from the other five newspapers.
Keywords-in-context analysis allowed me to have a quick understanding of the contextual meaning of the term latinx according to each newspaper. I found that identity and political participation/representation were important contexts to the latinx term for the Washington Post and Miami Herald; while cultural production and representation within the moving image industry were important contexts for Los Angeles Times and The New York Times; for the Houston Chronicle on the other hand, college life and identity were important contexts, while religion was a distinctive context for the latinx term only present in news articles from Chicago Tribune.
CONCLUSION
Here I applied computational analysis to news articles searched on the term latinx from six newspapers in order to compare them according to (1) Latin American countries mentioned, (2) their written topics, and (3) the word context in which the latinx term was mentioned. I found three countries were preferentially cited (Cuba, Mexico and Puerto Rico), and these mentions varied according to the newspaper. This result suggests that latinx might not be a comprehensive term in relation to representation of people from many other Latin American countries living in United States. Latinx for the Washington Post and Miami Herald probably refers to Cubans, whereas for The New York Times and Chicago Tribune it refers to Puerto Ricans, and for Los Angeles Times and Houston Chronicle it mainly refers to Mexicans. Using machine learning approaches, I extracted six topics and evaluated their relative proportions according to each newspaper. Topic number 2, which seemed to be related to political participation/representation was most prominent in the Miami Herald, whereas topic 1 ,which seemed to be related health and housing related issues during the pandemic, was most prominent in the Chicago Tribune. The word contexts surrounding the term latinx also differed depending on the newspaper in question. Identity and political representation seemed to be relevant for The Washington Post and Miami Herald, whereas cultural production seemed to be relevant for Los Angeles Times and The New York Time; with religion being relevant for the Chicago Tribune. My work sheds lights on how the printed press in the United States viewed the term latinx and how this view differed among newspapers. This is an important step towards a better understanding of a new term.
REFERENCES
ABOUT THE AUTHOR
Martin Calvino is a Visiting Professor at Torcuato Di Tella University; a Computational Biologist at The Human Genetics Institute of New Jersey — Rutgers University; and a Multimedia Artist. You can follow him on Instagram: @from.data.to.art
THE CODE FOR THIS WORK CAN BE ACCESSED AT MY GITHUB PAGE:
