Generating static linguistic motion charts

1 Introduction

This document shares our R codes to generate publication-ready, static linguistic motion charts. The codes were initially designed to create “Figure 2” in our paper (Primahadi-Wijaya-R. & Rajeg, 2014) on the distributional change of nominal collocates for hot and warm. That paper was inspired by a pioneering study by Martin Hilpert (cf. Hilpert, 2011), who is the first to introduce motion charts to diachronic linguistics as tools for visualising the dynamic process of language change. The static version of the charts represents sequentially ordered bubble/scatter plots (Figure 2.1) of certain linguistic phenomena across different diachronic corpus periods on the basis of the Corpus of Historical American English (COHA) (Davies, 2012). We learnt the R function (i.e., the base R symbols() (cf., 2.1)) to generate the bubble(/scatter) plot in our paper from section 11.1.4 in Kabacoff’s (2011, pp. 278–279) book about the “bubble plot”.

There have been two accessible tutorials on generating linguistic motion charts. One is from Martin Hilpert himself via his YouTube demo and tutorials to generate the dynamic/interactive version of the motion charts using the googleVis R package (Gesmann & de Castillo, 2011). The second one is by Desagulier (2016), who demonstrates how one can retrieve the data from COHA and wrangle it in R into the required input format for the gvisMotionChart() function in the googleVis package. The present document aims to complement these with our R codes to create the static version of the motion charts.

We present here two flavours of the codes: (i) the base R (with minor tweak from the 2014 version used in our paper [and in Pichler (2016)]) (2.1); and (ii) the ggplot2 version (2.2). We first illustrate the codes using our hotwarm.RData available for download. Then, we re-use the (much simpler) ggplot2 codes to reproduce the static motion chart in Hilpert’s (2011, p. 455, Figure 8) case study 2 on the English complement-taking predicates, testing the potential reproducibility of the ggplot2 version (2.3) (but, readers are encouraged to play around with the base R version).

2 Main steps

Once the hotwarm.RData file has been downloaded, the easiest way to load the data is by double-clicking the file to open an R session. Alternatively, one may load it from an R console using load(), provided that the .RData is stored in the same working directory with the current R session.

load(file = "hotwarm.RData")

The R workspace consists of a data frame called hotwarm with the following format.

head(hotwarm)

##   decade       word        hot      warm      freq diff
## 1   1860    admirer 0.00000000 0.4087567 0.4087567 11.0
## 2   1860  affection 0.00000000 0.9343010 0.9343010 25.2
## 3   1860  afternoon 0.23357525 0.2919691 0.5255443  1.0
## 4   1860        air 0.64233194 1.0510886 1.6934206  1.3
## 5   1860 attachment 0.00000000 0.4087567 0.4087567 11.0
## 6   1860        bed 0.05839381 0.5255443 0.5839381  7.1

The decade represents the COHA decades. In the paper, we track the nominal collocational change of hot and warm across fifteen decades, from the 1860s to the 2000s. The hot and warm columns show the normalised co-occurrence frequencies per million words of the nouns (in the word column) with each hot and warm respectively in the [adj + Noun] schema. In other words, we retrieved the R1 collocates of the two adjectives from COHA. The freq column contains the joint frequency of the nouns in the specified schema across hot and warm (i.e., the sum of values in hot and warm). Finally, the diff column uses the values from the “SCORE” column produced by the “compare” feature of COHA (this choice follows the suggestion in Martin Hilpert’s tutorial).

2.1 Base R version of the static motion chart

The steps for generating the static chart with the base R are described below.

1. Generate character vectors for the labels of each decade.

decade.label <- as.character(unique(hotwarm$decade))
decade.label

##  [1] "1860" "1870" "1880" "1890" "1900" "1910" "1920" "1930" "1940" "1950"
## [11] "1960" "1970" "1980" "1990" "2000"

2. Create a character vector containing collocates whose distribution with hot and warm will be traced through. This vector will be used as the selected labels for the bubbles in the scatter plots.

selected.collocs <- c("bath","day","dog","heart","pursuit","smile","spot","water","welcome")

3. Subset the data points for the selected collocate-labels defined in step (2) from the hotwarm data frame.

dframe.label <- subset(hotwarm, word %in% selected.collocs)
head(dframe.label)

##    decade    word        hot      warm      freq diff
## 16   1860     day 0.99269482 1.3430577 2.3357525  1.1
## 26   1860   heart 0.05839381 2.3941463 2.4525401 32.2
## 32   1860 pursuit 0.81751338 0.0000000 0.8175134 35.6
## 45   1860   water 5.89777509 1.1094824 7.0072575  6.8
## 47   1860 welcome 0.00000000 0.5255443 0.5255443 14.2
## 54   1870    bath 0.21493636 1.5582886 1.7732249  5.6

4. Split the hotwarm data frame into a list of subset data frame by decades.

# data frame for all data points
df.main <- split(hotwarm, hotwarm$decade) 

# check the type of the data and the length
typeof(df.main)

## [1] "list"

length(df.main)

## [1] 15

5. Split also the dframe.label produced in (3) above into a list of data frame per decade, containing the data for the collocates labels.

# data frame for the selected collocates
df.labs <- split(dframe.label, dframe.label$decade) 

6. Now comes the for loop to generate the scatter plots for each of the fifteen decades. Readers might do some adjustment from the codes below, such as the limits, or tick-labels, of the x- (xlim) and y-axis (ylim), given the nature of their data. Before running the for loop, we need to define a number of graphical parameters for the ploting (with par(), cf. 6.1 in the chunk below), especially for the number of row and column in the plot. The hotwarm data consists of fifteen decades. So we divide the plot into 5-by-3 matrix (hence, mfrow = c(5, 3)). Details for the list of arguments in the par() call below can be reviewed by typing ?par() in the console.

# 6.1 Define plotting parameter---------
par(mfrow = c(5, 3), # set a 5-by-3 plotting space
    cex = 0.425, 
    mar = c(2.5, 2, 1, 1), 
    mgp = c(3, 0.75, 0), 
    tck = -0.025, 
    las = 1) 

# 6.2 "for" loop to generate the sequential scatterplots---------
for (i in seq_along(decade.label)) { # loop sequence is derived from the total number of (15) decades
    
  # preparing empty plot field from the dataset of all decades
    plot(df.main[[i]][,3], # column [,3] for the x-axis
         df.main[[i]][,4], # column [,4] for the y-axis
         type = "n", 
         xlim = c(-0.5, max(hotwarm$hot)+1), 
         ylim = c(-0.5, max(hotwarm$warm)+0.5), 
         yaxt = "n", 
         xaxt = "n", 
         ann = F) 
    
  # adding major y-axis to the empty plot
    axis(side = 2, at = c(0, 1, 2, 3), labels = c(0, 1, 2, 3)) 
    
  # (OPTIONAL) adding minor y-axis to the empty plot
    #axis(side = 2, at = c(1, 3), labels = FALSE, tcl = -0.2) 
    
  # adding major x-axis to the empty plot
    axis(side = 1, at = c(0, 2, 4, 6, 8), labels = c(0, 2, 4, 6, 8), tick = TRUE) 
    
  # (OPTIONAL) adding minor x-axis to the empty plot
    #axis(side = 1, at = c(1, 3, 5, 7), labels = FALSE, tcl = -0.2) 
    
  # adding gridlines onto the empty plot field
    grid(lwd = 1.15) 
    
  # adding the decade labels onto the plot
    text(4, 3, labels = decade.label[i], col = "grey85", cex = 4.15, font = 2) 
    
  # adding the bubble symbol onto the plot from all data points (cf. Kabacoff, 2011, pp. 278-279)
    symbols(df.main[[i]][,3], 
            df.main[[i]][,4], 
            inches = 0.150, 
            fg = "white", 
            bg = "grey85", 
            circles = sqrt(df.main[[i]][,5]/pi), # column [,5] for the "freq" used as bubble-size input
            add = TRUE) 
    
  # adding the bubble symbol onto the plot with darker colour for the labelled data points  
    symbols(df.labs[[i]][,3], 
            df.labs[[i]][,4], 
            inches = 0.150, 
            fg = "white", 
            bg = "grey65", 
            circles = sqrt(df.labs[[i]][,5]/pi), 
            add = TRUE) 
    
  # adding the selected collocate labels onto the plot
    text(df.labs[[i]][,3], 
         df.labs[[i]][,4], 
         labels = df.labs[[i]][,2], 
         cex = 1.25) 
    
    } # end of "for-loop"

Figure 2.1: Changes in the collocational profiles of hot (x-axis) and warm (y-axis), COHA 1860s-2000s (Primahadi-Wijaya-R. & Rajeg, 2014, p. 253).

In order to save the above plot, first set up a .png file (or other formats, such as .jpeg or .pdf) for the resulting plot, before executing the above codes, and define the plot’s resolution and sizes. After that, run the previous codes in 6.1 and 6.2 (i.e., step (6)), and end with dev.off() to close the graphic device and save the plot. See the code-chunk below.

# set up a .png file for the resulting plot, including the plot's resolution and sizes
png("hotwarmPLOT-loop.png", 
    res = 450, 
    width = 6.5 * 500, 
    height = 8.5 * 500)

# Insert here (and then run) the (i) plotting parameter call (6.1) and (ii) the for loop (6.2)

# close the active graphic device to save the plot
dev.off ()

Note that to achieve the desired results in terms of the size and resolution of the plot, a couple of trial-and-errors should be expected.

2.2 The ggplot2 version of the static motion chart

The ggplot2 package offers a facetting feature that is the key to a (much) simpler code for producing the static motion charts. In ggplot2, we can use the facet_wrap() function, which can be directly fed with the decade column and let ggplot2 do the job to split the plots by decades. To run the following codes, readers need to either install ggplot2 package separately or together as a part of a suite of packages called the tidyverse (Wickham & Grolemund, 2017). Optionally, readers might also want to install the ggrepel package that is used here to handle the overlapping labels.

Here are the steps for the ggplot2 version.

1. Prepare two data frames: (i) one for the selected collocate/item labels (df.label), and (ii) one for the remaining items (df.others), which will not be labelled in the plot. Here, we also generate the axes labels.

# get the data for the selected collocates
df.label <- subset(hotwarm, word %in% selected.collocs) 

# get the data for the remaining collocates
df.others <- subset(hotwarm, !word %in% selected.collocs)

# x-axis labels
xaxis <- expression(paste("Co-occurrence frequency per million words with ", italic("hot"), sep = ""))

# y-axis labels
yaxis <- expression(paste("Co-occurrence frequency per million words with ", italic("warm"), sep = ""))

2. Generate the first ggplot() call for the non-labelled data and save the output into p.

# load the ggplot2 via tidyverse (or individually) and load the ggrepel
library(ggplot2)
library(ggrepel)

# generate plot for the non-labelled data
p <- ggplot(data = df.others,
            aes(x = hot, y = warm)) +
  geom_point(aes(size = freq), 
             alpha = 1/8, 
             show.legend = FALSE) +
  facet_wrap(~decade) # here is the key to split the plot by decades!

3. Add the labelled data points to the previous plot (p) on the basis of the df.label data frame.

p <- p + geom_point(data = df.label,
               aes(size = freq),
               alpha = 1/2,
               colour = "royalblue",
               show.legend = FALSE) +
  # if you are not using `ggrepel` below, use the standard `geom_text()` function
  geom_text_repel(data = df.label,
                  aes(label = word), 
                  size = 2.5) +
  labs(x = xaxis, # insert the `xaxis` data
       y = yaxis) + # insert the `yaxis` data
  theme_bw()

# call out the plot
p

Figure 2.2: Changes in the nominal collocational profiles of hot and warm, COHA 1860s-2000s (Primahadi-Wijaya-R. & Rajeg, 2014, p. 253).

4. Save the plot as .png file using ggsave().

ggsave(file = "hotwarm.png", plot = p, width = 20, height = 20, unit = "cm", dpi = 600)

Note that trial and errors are expected to get the desired results in terms of the size and resolution. For the previous plot, the parameters for width, height, and resolution (i.e., dpi) given above produce a desirable result from our own perspective.

2.3 Recreating previous motion chart in Hilpert (2011)

To recreate the chart (i.e., “Figure 8”) in Hilpert’s (2011) paper, we need to download the data from Martin Hilpert’s motion charts resource page. The data is available as zip file, which contains the mot.RData workspace. After the downloads, open/load the R workspace that contains two data frames: (i) convdata and (ii) compdata. The former is the data for Hilpert’s (2011, p. 444) case study 1 on the “ambicategorical nouns and verbs”. We will instead use the compdata to produce the static motion charts for the English complement-taking predicates (readers can try the codes for themselves with the convdata).

load("mot.RData")
head(compdata)

##             x          y decade        verb       freq
## 1 -0.29379995  0.1350332   1860 acknowledge  12.583587
## 2 -0.38929710  0.2318209   1860       admit  34.346505
## 3 -0.54856108  0.3116948   1860      affirm   8.206687
## 4  0.01545954 -0.6818823   1860  appreciate   1.519757
## 5  0.08741028 -0.7591102   1860       await   5.957447
## 6 -0.28720186  0.2074660   1860     believe 178.115502

The x- and y-axis represent the dimensions of the metric Multidimensional Scaling (MDS) from the (dis)similarity measures between the verbs on the basis of their complementation profiles (Hilpert, 2011, pp. 451–452, 454). The freq column indicates the normalised text frequencies of the verbs (Hilpert, 2011, p. 452). The steps to generate the ggplot2 motion charts from the compdata data are as follows.

1. As above, define two data frames, one for the labelled items and one for the remainder of the data points. We also store the labels for the axes (cf., Hilpert, 2011, p. 453, for the interpretation given on the x- and y-axes).

# selected complement-taking predicates
predicates <- c("confirm", "await", "enjoy", "demand", "consider", "remember", "hope", "dislike", "want", "try")

# get the data for the selected predicates
df.label <- subset(compdata, verb %in% predicates) 

# get the data for the remaining predicates
df.others <- subset(compdata, !verb %in% predicates)

# xaxis label
xaxis <- expression(paste(italic("that"), "-clauses <----------    ----------> ", italic("to"), "-infinitives", sep = ""))

# yaxis label
yaxis <- "nominal compl. <----------    ----------> complex verbal/clausal compl."

2. Generate the first ggplot() call for the non-labelled data and save the output into p.

# generate plot for the non-labelled data
p <- ggplot(data = df.others,
            aes(x = x, y = y)) +
  geom_point(aes(size = freq), 
             alpha = 1/8, 
             show.legend = FALSE) +
  facet_wrap(~decade) # here is the key to split the plot by decades!

3. Add the labelled data points to the previous plot (p) on the basis of the df.label data frame.

p <- p + geom_point(data = df.label,
               aes(size = freq),
               alpha = 1/2,
               colour = "royalblue",
               show.legend = FALSE) +
  # if you are not using `ggrepel` below, use the standard `geom_text()` function
  geom_text_repel(data = df.label,
                  aes(label = verb), 
                  size = 2.5) +
  labs(x = xaxis, # insert the `xaxis` data
       y = yaxis) + # insert the `yaxis` data
  theme_bw()

# call out the plot
p

Figure 2.3: Two-dimensional MDS solution for the English complement-taking predicates, COHA 1860s-2000s (Hilpert, 2011, p. 455, Figure 8).

We can save the plot using the ggsave() function following the code in section 2.2 with the hotwarm data.

3 Coda

We hope this document can be useful for those interested in creating other motion charts, especially the static ones. The .Rmd source file of this document, including the .bib and csl files, are available in our figshare accounts (cf. here and/or here). This HTML document is rendered with Xie’s (2016) bookdown R package in RStudio. For any feedback and comments, please drop us email at primahadiwijaya@gmail.com. We would be happy to hear from you!

# session info for this R document
# just updating to R 3.4.3 and the codes still work!
> devtools::session_info()
Session info ------------------------------------------------------------------
 setting  value                       
 version  R version 3.4.3 (2017-11-30)
 system   x86_64, darwin15.6.0        
 ui       RStudio (1.1.383)           
 language (EN)                        
 collate  en_US.UTF-8                 
 tz       Australia/Melbourne         
 date     2018-02-09                  

Packages ----------------------------------------------------------------------
 package    * version date       source        
 base       * 3.4.3   2017-12-07 local         
 colorspace   1.3-2   2016-12-14 CRAN (R 3.4.0)
 compiler     3.4.3   2017-12-07 local         
 datasets   * 3.4.3   2017-12-07 local         
 devtools     1.13.4  2017-11-09 CRAN (R 3.4.2)
 digest       0.6.15  2018-01-28 CRAN (R 3.4.3)
 ggplot2    * 2.2.1   2016-12-30 CRAN (R 3.4.0)
 ggrepel    * 0.7.0   2017-09-29 CRAN (R 3.4.2)
 graphics   * 3.4.3   2017-12-07 local         
 grDevices  * 3.4.3   2017-12-07 local         
 grid         3.4.3   2017-12-07 local         
 gtable       0.2.0   2016-02-26 CRAN (R 3.4.0)
 knitr        1.19    2018-01-29 CRAN (R 3.4.3)
 labeling     0.3     2014-08-23 CRAN (R 3.4.0)
 lazyeval     0.2.1   2017-10-29 CRAN (R 3.4.2)
 memoise      1.1.0   2017-04-21 CRAN (R 3.4.0)
 methods    * 3.4.3   2017-12-07 local         
 munsell      0.4.3   2016-02-13 CRAN (R 3.4.0)
 pillar       1.1.0   2018-01-14 CRAN (R 3.4.3)
 plyr         1.8.4   2016-06-08 CRAN (R 3.4.0)
 Rcpp         0.12.15 2018-01-20 CRAN (R 3.4.3)
 rlang        0.1.6   2017-12-21 CRAN (R 3.4.3)
 scales       0.5.0   2017-08-24 CRAN (R 3.4.1)
 stats      * 3.4.3   2017-12-07 local         
 tibble       1.4.2   2018-01-22 CRAN (R 3.4.3)
 tools        3.4.3   2017-12-07 local         
 utils      * 3.4.3   2017-12-07 local         
 withr        2.1.1   2017-12-19 CRAN (R 3.4.3)
 yaml         2.1.16  2017-12-12 CRAN (R 3.4.3) 

References

Davies, M. (2012). Expanding horizons in historical linguistics with the 400-million word Corpus of Historical American English. Corpora, 7(2), 121–157.

Desagulier, G. (2016). A motion chart of Captain Kirk’s split infinitive with R and googleVis. Retrieved January 31, 2018, from http://rstudio-pubs-static.s3.amazonaws.com/153634_658c5b5686cd43e591c4e31c0437f724.html

Gesmann, M., & de Castillo, D. (2011). GoogleVis: Interface between r and the google visualisation api. The R Journal, 3(2), 40–44. Retrieved from https://journal.r-project.org/archive/2011-2/RJournal_2011-2_Gesmann+de~Castillo.pdf

Hilpert, M. (2011). Dynamic visualizations of language change. Motion charts on the basis of bivariate and multivariate data from diachronic corpora. International Journal of Corpus Linguistics, 16(4), 435–461.

Kabacoff, R. I. (2011). R in Action. Data analysis and graphics with R. Shelter Island, New York: Manning.

Pichler, K. (2016). A diachronic perspective on synonymy (Diploma Thesis). University of Vienna, Vienna, Austria. Retrieved from http://othes.univie.ac.at/40365/

Primahadi-Wijaya-R., G., & Rajeg, I. M. (2014). Visualising diachronic change in the collocational profiles of lexical near-synonyms. In I. N. Sudipa & G. Primahadi-Wijaya-R. (Eds.), Cahaya Bahasa: A festschrift in honour of Prof. I Gusti Made Sutjaja (pp. 247–258). Denpasar: Swasta Nulus. Retrieved from http://doi.org/10.4225/03/564A9F35BEDB4

Wickham, H., & Grolemund, G. (2017). R for Data Science. Canada: O’Reilly. Retrieved from http://r4ds.had.co.nz/

Xie, Y. (2016). Bookdown: Authoring Books and Technical Documents with R Markdown. Chapman and Hall/CRC.