In csqsiew/netlangr: Generates Language Networks and Network Measures

This vignette demonstrates how to use the main functions in the netlangr package.

The makelangnet and makemultinet functions are used to create a language network from a list of words.

The getnetstats and getmultistats functions are used to compute the network measures from a language network.

The makelangnet and getnetstats functions are for single-layer networks, i.e., either phonological or orthographic similarity networks.

The makemultinet and getmultistats functions are for multi-layer networks, i.e., the phono-graphic multiplex where both phonological and orthographic links are represented in the network.

The package is still very much a work in progress, so any feedback, comments, and suggestions are very welcome!

Set up

Download the package from my github page.

# install.packages('devtools') 
# library(devtools)
# install_github('csqsiew/netlangr')
library(netlangr)

Load some example data

All you really need to get started is a list of words. Spellings if you want to construct an orthographic network, and phonological transcriptions if a phonological network is desired. Note that the phonological transcriptions must be constructed such that 1 phoneme = 1 character, this is because the networks are constructed based on edit distance of 1 (i.e., a link is placed between pairs of words that differ by the substitution, deletion, or addition of one phoneme/letter, which is the way that phonological or orthographic similarity is typically operationalized in the psycholinguistic literature; Luce & Pisoni, 1998; Coltheart et al., 1977).

data <- read.csv('cat.csv', stringsAsFactors = F) 
# stringsasFactor = F to force the columns to character class, instead of factor
head(data)
class(data$Phono) # should be 'character'
class(data$Ortho) # should be 'character'

Single-layered network

# Phonological network
phono.net <- makelangnet(data$Phono) # make the language network 
phono.net.measures <- getnetstats(phono.net) # get network measures 
head(phono.net.measures)

# Orthographic network
ortho.net <- makelangnet(data$Ortho) # make the language network 
ortho.net.measures <- getnetstats(ortho.net) # get network measures 
head(ortho.net.measures)

Network measures:

location: G = largest connected component (giant component), L = lexical island, H = hermit

degree: number of words that are neighbors of a given word (i.e., neighborhood size)

clustering: the extent to which a word's neighbors are also neighbors of each other, i.e., clustering in a word's neighborhood in the network, ranges from 0 to 1.

closeness.gc: normalized inverse of the average distance between a given word and all other words in the LCC, higher values indicate that a word is close to many other words in the network (more central). (Note that closeness centrality is only calculated for words in the LCC.)

Multi-layered network

multi.net <- makemultinet(data)
multi.net.measures <- getmultistats(multi.net)
head(multi.net.measures)
# write.csv(multi.net.measures, file = 'output.csv') # export the data if you wish

Network measures:

location: G = largest connected component (giant component), L = lexical island, H = hermit

degree.pg: number of words that are both phonologial AND orthographic neighbors of a given word (i.e., phonographic neighbors; the neighborhood size of the phonographic network)

degree.all: number of words that are phonological or orthographic neighbors of a given word (note that phonographic neighbors are not double counted, i.e., the neighborhood size of the phonographic multiplex)

clustering.pg: the extent to which a word's phonographic neighbors are also phonographic neighbors of each other, i.e., clustering in a word's neighborhood in the phonographic network, ranges from 0 to 1.

clustering.unweighted: the extent to which a word's phonological and orthographic neighbors are phonological or orthographic neighbors of each other, i.e., clustering in a word's neighborhood in the phonographic multiplex, ranges from 0 to 1. unweighted = each link has the same weight.

clustering.weighted: the extent to which a word's phonological and orthographic neighbors are phonological or orthographic neighbors of each other, i.e., clustering in a word's neighborhood in the phonographic multiplex, ranges from 0 to 1. weighted = phonographic links are double weighted as compared to phonological or orthogrpahic only links.

closeness.gc.unweighted: normalized inverse of the average distance between a given word and all other words in the LCC, higher values indicate that a word is close to many other words in the network (more central). unweighted = each link has the same weight.

closeness.gc.weighted: normalized inverse of the average distance between a given word and all other words in the LCC, higher values indicate that a word is close to many other words in the network (more central). weighted = phonographic links are double weighted as compared to phonological or orthogrpahic only links. (Note that closeness centrality is only calculated for words in the LCC.)

A pretty network figure

library(igraph)

l <- layout_with_lgl(multi.net)

# color edges by their connection type 
E(multi.net)$color <- E(multi.net)$type
E(multi.net)$color <- E(multi.net)$color %>% gsub('po', 'green', .) %>% 
  gsub('o', 'blue', .) %>% gsub('p', 'red', .)

plot(multi.net, 
     vertex.label.color = 'black',
     vertex.color = 'white',
     vertex.label.family = 'Helvetica',
     layout = l,
     edge.color = E(multi.net)$color,
     vertex.label.cex = .7,
     vertex.shape="none", vertex.label=V(multi.net)$label,
     main = 'Phonographic network of CAT')

legend(x=-1.5, y=-1.1, c("phonographic","orthographic", "phonological"), pch=21,
       col="#777777", pt.bg=c('green', 'blue', 'red'), pt.cex=2, cex=.8, bty="n", ncol=1)

References

Phonological clustering coefficient in SWR: Chan & Vitevitch (2009)
Closeness centrality in SWR: Goldstein & Vitevitch (2017)
Network components in SWR: Siew & Vitevitch (2016)
Small world networks, clustering coefficient: Watt & Strogatz (1998)
Centrality measures: Borgatti & Everett (2006)
Multilayer networks: Boccaletti et al., (2014), Menichetti et al., (2014)
Neighborhood density/degree: Luce & Pisoni (1998), Coltheart et al. (1977), Andrews (1997)
Phonological language network: Vitevitch (2008)