Interactive language maps in R with lingtypology

67 min read R, languistics, maps


The awesome lingtypology package by George Moroz can be used for many purposes. In this tutorial, we’ll explore various features showing how the package can be used to a) gather data from Glottolog about the many languages of the world; b) showcase your research with interactive maps and clickable content, and c) create teaching materials.

Contents

  1. Installing and loading lingtypology
  2. Where is a particular language spoken?
  3. Which languages are spoken in a given country?
  4. Gathering data about a language
  5. Creating a data frame with all data for a specific country
  6. Mapping the languages of a given country
  7. Mapping a set of languages with custom features
  8. Mapping a language using custom coordinates
  9. Adding a pop-up video to a map
  10. Changing the map type
  11. Mapping members of a language family
  12. Mapping two language families
  13. Getting data from typological databases
  14. Mapping predictions about linguistic features
  15. Mapping links between languages on an interactive globe

1. Installing and loading lingtypology

if (!require('lingtypology')) {
  install.packages("lingtypology")}
library(lingtypology)

2. Where is a particular language spoken?

Let’s start with something simple, and ask where, say, Michif is spoken.

country.lang("Michif")
##                  Michif 
## "Canada, United States"

3. Which languages are spoken in a given country?

We can also ask start with a country and ask which languages are spoken in it. Let’s try Australia and save the output in a variable we call ausLang.

ausLang = lang.country("Australia")
length(ausLang) # returns the number of languages in our vector
## [1] 421
head(ausLang) # returns the first few entries
## [1] "Southern Coastal Yuin"               
## [2] "Tasmanian"                           
## [3] "New South Wales Pidgin"              
## [4] "Uwinymil"                            
## [5] "Senaya"                              
## [6] "Oyster Bay-Big River-Little Swanport"

4. Gathering data about a language

We can also get various data about a language. Let’s see what we can learn about Gooniyandi.

gooniyandi = data.frame(subset(glottolog.original,language=="Gooniyandi"))
names(gooniyandi) # lists the variables in the data frame
##  [1] "language"           "iso"                "glottocode"        
##  [4] "longitude"          "latitude"           "affiliation"       
##  [7] "area"               "alternate.names"    "affiliation.HH"    
## [10] "country"            "dialects"           "language.status"   
## [13] "language.use"       "location"           "population.numeric"
## [16] "typology"           "writing"

5. Creating a data frame with data for a specific country

We also remove rows in which all data columns are “NA” and limit our data frame to include the following variables:
1. language
2. family
3. location
4. language status

Our data frame will be nice and tidy, but due to long strings within certain variables, it’s difficult to inspect our data. We therefore write our data to a html table using xtable(). However, you need to uncomment (remove the “#” symbol) the final line.

za = data.frame(glottolog.original)
za = za[za$country=="South Africa",]
ind <- apply(za, 1, function(x) all(is.na(x)))
za <- za[ !ind, ]
za=za[,c(1,6,12,14)] # the column indices corresponding to the variables we want

if (!require('xtable')) 
install.packages("xtable")
## Loading required package: xtable

library(xtable)
za=xtable(za)
#print.xtable(za, type="html", file="za1.html")

6. Mapping the languages of a given country

We use our previously declared variable ‘ausLang’ to create an interactive map using the map.feature function. Alternatively, we could simply use map.feature(lang.country(“Australia”)).

map.feature(ausLang)
## Warning: There is no coordinates for languages Southern Coastal Yuin,
## Tasmanian, New South Wales Pidgin, Uwinymil, Wathawurrung, Kuuk-Yak, Lower
## Southern Aranda, Barranbinya, Ngadjuri, Worrorra, Djiwarli, Kawarrang-
## Ogh Undjan, Birrdhawal, Karruwali, Kabikabi, Ganai, Yadhaykenu, Bindal-
## Cunningham, Dharumbal, Guwa, Unggumi, Wirangu, Yaygir, South Australian
## Pidgin English, Guyambal, Ngumbarl, Wuthathi, Aritinngitigh, Bindal-Mount
## Elliot, Kalaamaya, Kuungkari of Barcoo River, Gulunggulu, Anguthimri,
## Koko Dhawa, Wakabunga, Giya, Pidgin Kaurna, Gugu Mini, Queensland Kanaka
## English, Koko Babangk, Ikaranggal, Bigambal, Bunganditj, Athima, Thiin,
## Lower Riverland, Sydney, Upper Riverland, Woiwurrung, Yulparija, Pidgin
## Ngarluma, Aghu Tharnggalu (Retired), Bindal-Gorton, Arakwal, Wulwulam,
## Tagalaka, Yuru, Pirriya, Anggamuthi, Marriammu, Light Warlpiri, Hawkesbury,
## Ngardi, Nyiyaparli, Olkol, Yanda, Yinhawangka, Yirandhali, Yugul

7. Mapping a set of languages with custom features

We can also produce a map of languages with user-specified features such as case-marking.

myLanguages=c("Nyulnyul", "Warrwa", "Guugu Yimidhirr","Warlpiri","Gooniyandi")
myFeatures=c("accusative","unknown","neutral","unknown","accusative")
map.feature(myLanguages,myFeatures)

8. Mapping a language using custom coordinates

Some languages appear to be missing geographical coordinates, which the user can then supply, or change cureent coordinates. In the following map, we will also add these features:
1. zoom control
2. zoom level
3. a minimap
4. pop-up text 

map.feature("Gooniyandi",
            label="Gooniyandi", 
            minimap=T, #logical value, True or False (or T/F), False by default
            zoom.control=T, 
            zoom.level=3,
            popup="You can add additional info here <br>another line with info",
            latitude = -19, 
            longitude = 125)

9. Adding a pop-up video to a map

We may even add a popup video ((which unfortunately might not show properly on this site). This feature could be useful for field linguists who would like to add a short video introduction to a language they’re documenting, maybe even from the field. The particular video in this example is unrelated to Gooniyandi and only intended to illustrate the idea.

video="https://media.spreadthesign.com/video/mp4/13/48600.mp4"
video= paste("<video width='200' height='150' controls> <source src='",
                         as.character(video),
                         "' type='video/mp4'></video>", sep = "")
map.feature("Gooniyandi",popup=video,zoom.level=4)

10. Changing the map type

There are several map types available. have a look here: https://leaflet-extras.github.io/leaflet-providers/preview/index.html

By adding the control parameter, we can also create a map with two layers to choose from.

map.feature("Swedish", tile =c("OpenTopoMap","Stamen.Watercolor"),control=T, zoom.level=4)


The world at night. This map and the next seem to not show unless we specify a zoom level. 

map.feature("Swedish", tile ="NASAGIBS.ViirsEarthAtNight2012",zoom.level=5)


My personal favourite.. 

map.feature("Swedish", tile ="Thunderforest.SpinalMap",zoom.level = 5)

11. Mapping members of a language family

We will map the Khoisan languages and add a density contour.

map.feature(lang.aff("Khoisan"),density.estimation = TRUE,density.width=5)


And if we only want the area without points.. 

map.feature(lang.aff("Khoisan"),density.estimation = TRUE,density.points = FALSE,density.width=5)

12. Mapping two language families

Mapping two language families gets slightly more tricky and requires a bit of coding. Here we’ll map Bantu and Khoisan languages.
1. We first gather the languages belonging to Khoisan and Bantu.
2. Because the aff.lang variable contains many details, we use grepl() to search the strings of text for “Bantu” and “Khoisan” and assign these labels in a new variable we call “family”.
3. We then join the individual languages with their family labels in a data frame.
4. As a bonus, we add coordinates to our data.
5. Lastly, we remove the redundant row names. 

language=lang.aff(c("Khoisan","Bantu")) # Step 1
family=aff.lang(language) # this is an extra step needed for this document to work
family[grepl("Bantu",aff.lang(language))==T]="Bantu" # Step 2
family[grepl("Khoisan",aff.lang(language))==T]="Khoisan"
africa=data.frame(language,family) # Step 3
africa$long=long.lang(africa$language) # Step 4
africa$lat=lat.lang(africa$language) 
rownames(africa) <- c() # Step 5
head(africa)
##           language  family     long       lat
## 1             Xiri Khoisan 20.72598 -28.42578
## 2          Sandawe Khoisan 35.48081  -5.26918
## 3          Hai//om Khoisan 17.02985 -19.76371
## 4             /Xam Khoisan 20.17325 -31.76115
## 5 North-Central Ju Khoisan 18.00000 -21.92000
## 6          //Xegwi Khoisan 30.40283 -26.34068
table(africa$family)
## 
##   Bantu Khoisan 
##     535      27

We can now plot our data. 

map.feature(africa$language,
            features=africa$family,
            longitude = africa$long,
            latitude = africa$lat,
            density.estimation = africa$family,
            density.width=5)
## Warning: There is no coordinates for languages Ndambomo, Marachi, Marama,
## Shiwa, Kempee, Nyika (Tanzania), Nyika (Malawi and Zambia), Hungu, Khayo,
## Wanga, Tunen (Retired), West Nyala, Nyiha (Malawi), Viya, Kabras, Osamayi,
## Ngubi, Mikaya-Bambengangale-Baluma, Belueli, Tachoni

We can also create the same map with a few modifications. 

map.feature(africa$language,
            features=africa$family,
            longitude = africa$long,
            latitude = africa$lat,
            density.estimation =africa$family,
            density.width=5,
            color=c("red","blue"),
            density.estimation.opacity=0.3,
            density.estimation.color = c("red","blue"),
            zoom.level=4,
            zoom.control = T)

13. Getting data from typological databases

We can download and use data from the following sources:
1. WALS
2. AUTOTYP
3. PHOIBLE
4. Affix borrowing database
5. South American indigenous language structures
6. Austronesian basic vocabulary database

As an example, lets map Matt Dryer’s basic word order data (WALS feature 81a).

wordOrder <- wals.feature(c("81a"))
## Don't forget to cite a source (modify in case of using individual chapters):
## 
## Dryer, Matthew S. & Haspelmath, Martin (eds.) 2013. The World Atlas of Language Structures Online. Leipzig: Max Planck Institute for Evolutionary Anthropology.
## (Available online at http://wals.info, Accessed on 2018-12-15.)
## 
## @book{wals,
##   address   = {Leipzig},
##   editor    = {Matthew S. Dryer and Martin Haspelmath},
##   publisher = {Max Planck Institute for Evolutionary Anthropology},
##   title     = {WALS Online},
##   url       = {http://wals.info/},
##   year      = {2013}
## }
head(wordOrder)
##   wals.code 81a  latitude longitude glottocode language
## 2       aba SOV  -4.00000    141.25   abau1245     Abau
## 3       abi SVO -29.00000    -61.00   abip1241   Abipon
## 4       abk SOV  43.08333     41.00   abkh1244   Abkhaz
## 5       abn SOV -28.25000    136.25   arab1267  Arabana
## 6       abo SOV   5.00000     36.75   arbo1245   Arbore
## 7       abu SVO  -0.50000    132.50   abun1252     Abun

map.feature(wordOrder$language,
            features = wordOrder$`81a`,
            latitude = wordOrder$latitude,
            longitude = wordOrder$longitude,
            label = wordOrder$language,
            title = "Word Order",
            control=T,
            zoom.control = T)

14. Mapping predictions about linguistic features

This is a very simple example of how you can map predictions about typological features based on a machine learning approach. In this case, we will use a decision tree for classification to predict word order purely from geographical coordinates. We could of course apply more sophisticated methods with model evaluation and include other predictor variables from e.g. WALS and Glottolog to optimize our model. However for our purposes, we will only consider longitude and latitude in a simple model.

library(rpart) # rpart is included in base R, but needs to be loaded

# We first merge our previously defined wordOrder data frame with Glottolog data.
# Our old wordOrder data frame is replaced by the new data frame
wordOrder=merge(glottolog.original,wordOrder,by="language")

# We then add a new column with word order data as a factor variable
wordOrder$wo=factor(wordOrder$`81a`)

# Now we can split our data into a training and a test set
set.seed(43282) #this allows for reproducibility of the analysis
smp_size <- round(0.7 * nrow(wordOrder)) #our training data will contain 70% of the rows
train_ind <- sample(seq_len(nrow(wordOrder)), size = smp_size) #choosing a random sample of row numbers
train <- wordOrder[train_ind, ]
test <- wordOrder[-train_ind, ] # the test set contains the rows not included in the training set

# We fit our simple model on the training set
fit <- rpart(wo ~ longitude.x+latitude.x,
             method="class", data=train,na.action = na.exclude)

# Based on our fitted model, we can assign the predicted word orders to a new variable 
# and add it to our test data frame
test$pred=predict(fit,test,type="class")

# We then create another column that tells us whether individual predictions were correct or not
# The logical operator '==' means "is equal to", compares two values and returns TRUE if values match
# '!=' means "is not equal to"
test$correct[test$pred==test$wo]="Correct"
test$correct[test$pred!=test$wo]="Incorrect"
test$correct=factor(test$correct)

# How many correct/incorrect predictions?
table(test$correct)
## 
##   Correct Incorrect 
##       278       122
cat("in percent: ", prop.table(table(test$correct))*100)
## in percent:  69.5 30.5

# Relative importance of our variables
fit$variable.importance
## longitude.x  latitude.x 
##    164.8253    116.3582
cat("Scaled to 100: ", fit$variable.importance/sum(fit$variable.importance)*100)
## Scaled to 100:  58.61841 41.38159

We now have a good idea about our model’s performance. Let’s create a map revealing correct and incorrect test predictions. Notice the text in the popups, which doesn’t seem to show clearly on this page.

map.feature(languages=test$language,
            features=test$correct,
            color=c("green","red"),
            minimap=T,
            zoom.control=T,
            popup = paste("actual: ",test$wo,"<br>","predicted: ",test$pred))
## Warning: There is no coordinates for languages Patwin
lingtypology
Avatar
Peer Christensen
Data Scientist
comments powered by Disqus