R Markdown Notebook for Vector Space Models and the usage patterns of Indonesian denominal verbs

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

1 How to cite this R Notebook

Please cite this R Notebook as follows (in Unified Style Sheet for Linguistics):

Rajeg, Gede Primahadi Wijaya, Karlina Denistia & Simon Musgrave. 2019. R Markdown Notebook for Vector space model and the usage patterns of Indonesian denominal verbs. figshare. https://doi.org10.6084/m9.figshare.9970205. https://figshare.com/articles/R_Markdown_Notebook_for_i_Vector_space_model_and_the_usage_patterns_of_Indonesian_denominal_verbs_i_/9970205.

2 Preface

# global option chunck
knitr::opts_chunk$set(fig.width = 7, 
                      fig.asp = 0.618,
                      dpi = 300,
                      dev = "pdf",
                      echo = TRUE,
                      message = FALSE,
                      warning = FALSE,
                      fig.path = "figures/")

This is an R Markdown Notebook (Rajeg, Denistia & Musgrave 2019a) for the statistical analyses accompanying our paper (Rajeg, Denistia & Musgrave 2019b) on vector space models and Indonesian denominal verbs (published in NUSA’s special issue titled Linguistic studies using large annotated corpora, edited by Hiroki Nomoto and David Moeljadi) (Nomoto & Moeljadi 2019). The Notebook, however, does not provide detailed exposition and discussion for each points, including English glossing of the Indonesian words in the tables. Readers are referred to our paper for details. Check the README page here for the tutorial to download this Notebook and the data (Rajeg, Denistia & Musgrave 2019c) for the paper.

The following R packages have to be installed and loaded to run all codes in this Notebook:

• cluster (Maechler et al. 2018)
• tidyverse (Wickham & Grolemund 2017)
• dendextend (Galili 2015)
• wordVectors (Schmidt & Li 2017)
• Rling (Levshina 2015)

Click the Code button to hide/reveal the R codes.

# load the required packages
## make sure all packages are installed!
library(cluster)
library(tidyverse)
library(dendextend)
library(wordVectors) # should be installed from Ben Schmidt's GitHub
library(Rling) # <- `Rling` package from Natalia Levshina's (2015) book 'How to do linguistics with R' published by John Benjamins.

3 Illustration for generating Vector Space Models (VSM)

For illustration, we used six deverbal nouns with the suffix -an. They are bacaan ‘reading’ (from the root baca ‘to read’), tulisan ‘writing’ (from tulis ‘to write’), lukisan ‘painting’ (from lukis ‘to paint/draw’), masakan ‘cooking’ (from masak ‘to cook’), makanan ‘food’ (from makan ‘to eat’), and minuman ‘beverages’ (from minum ‘to drink’). We retrieved collocates within three-word window to the either sides (left and right) of each noun. We used one of the Indonesian Leipzig Corpora files, namely ind_mixed_2012_1M-sentences.txt (see Biemann et al. 2007; Quasthoff & Goldhahn 2013).

3.1 Creating the co-occurrence frequency table

The following codes process the retrieved collocates into co-occurrence frequency table and print a subset of this table. The collocates data is available in the "vsm_creation_data.rds" file. This .rds file contains a List data-structure; the table version is available as tab-delimited .csv and .txt files (vsm_creation_data_tb.txt & vsm_creation_data_tb.csv).

df_list <- readRDS(file = "data/vsm_creation_data.rds")
df_ex <- purrr::map_df(df_list, dplyr::bind_rows); rm(df_list)
df_ex1 <- dplyr::filter(df_ex, 
                        !is.na(w), # remove NAs
                        nchar(w) > 1, # remove one-character tokens
                        stringr::str_detect(w, "^[a-z-]+$"),
                        
                        # remove incomplete words
                        !stringr::str_detect(w, "^[a-z]+-$"), 
                        
                        # remove incomplete words
                        !stringr::str_detect(w, "^-[a-z]+$")) 

df_ex1_count <- dplyr::count(df_ex1, node, w, sort = TRUE)
df_ex1_count_spread <- tidyr::spread(df_ex1_count, w, n, fill = 0)
df_ex1_count_spread <- as.data.frame(df_ex1_count_spread)
df_ex1_count_spread <- df_ex1_count_spread[, colSums(df_ex1_count_spread[, -1]) > 0]
rownames(df_ex1_count_spread) <- df_ex1_count_spread$node
df_ex1_count_spread <- df_ex1_count_spread[, -1]

selected_column <- c("alkitab", 
                     "dituangkan", "halal", "keras", 
                     "mengandung", "penutup")
df_ex1_print <- df_ex1_count_spread[, selected_column]
rownames(df_ex1_print) <- paste("*", rownames(df_ex1_print), "*", sep = "")
colnames(df_ex1_print) <- paste("*", colnames(df_ex1_print), "*", sep = "")
df_ex1_print$`...` <- rep("...", nrow(df_ex1_print))
knitr::kable(df_ex1_print, caption = "Raw co-occurrence frequency for the target words with the context words")

	alkitab	dituangkan	halal	keras	mengandung	penutup	…
bacaan	31	0	0	0	0	0	…
lukisan	0	0	0	1	0	0	…
makanan	0	0	19	12	55	7	…
masakan	0	0	0	0	1	0	…
minuman	0	0	0	150	14	0	…
tulisan	16	6	0	2	2	0	…

3.2 Performing the Positive Pointwise Mutual Information

Current approach in VSM adopts a more principled method of weighting the initial raw-frequency vectors into statistical measures of collocation strength before computing (dis)similarity measure between the target words. The goal of the weighting is “to give a higher weight to context words that co-occur significantly more often than expected by chance” (Heylen et al. 2015:156; cf. Clark 2015:503–504; Perek 2016:12). These significantly associatied context words are assumed to be more informative for the semantics of the target words (Heylen et al. 2015:156). The popular weighting measure used in VSM is the Pointwise Mutual Information (PMI) (see Levshina 2015:327–328 for computing PMI in R):

\[\text{PMI } (x, y) = log_{2} \frac {O_{xy}} {E_{xy}}\]

where \(O_{xy}\) represents the observed co-occurrence frequency between x and y, while \(E_{xy}\) is their expected co-occurrence frequency, which is the frequency expected under the chance distribution between x and y given the overall distribution of x and y in the corpus. Negative PMI values is normally replaced with zero, resulting in the Positive PMI (PPMI) (Levshina 2015; Hilpert & Perek 2015).

mtx <- as.matrix(df_ex1_count_spread)
mtx_exp <- chisq.test(mtx)$expected
mtx_pmi <- log2(mtx/mtx_exp); rm(mtx)
mtx_ppmi <- ifelse(mtx_pmi < 0, 0, mtx_pmi); rm(mtx_exp)
ppmiprint <- as.data.frame(round(mtx_ppmi, digits = 2)); rm(mtx_pmi)
ppmiprint <- ppmiprint[, selected_column]
rownames(ppmiprint) <- paste("*", rownames(ppmiprint), "*", sep = "")
colnames(ppmiprint) <- paste("*", colnames(ppmiprint), "*", sep = "")
ppmiprint$`...` <- rep("...", nrow(df_ex1_count_spread))
knitr::kable(ppmiprint, caption = "Weighted co-occurrence frequency with *Positive Pointwise Mutual Information*")

	alkitab	dituangkan	halal	keras	mengandung	penutup	…
bacaan	3.62	0.00	0.00	0.00	0.00	0.00	…
lukisan	0.00	0.00	0.00	0.00	0.00	0.00	…
makanan	0.00	0.00	1.05	0.00	0.67	1.05	…
masakan	0.00	0.00	0.00	0.00	0.00	0.00	…
minuman	0.00	0.00	0.00	3.07	0.85	0.00	…
tulisan	0.31	1.87	0.00	0.00	0.00	0.00	…

Further analysis can be performed. The most common one is determining the semantic (dis)similarity between the target words (i.e., which word is more similar and different among each other). The following section briefly discusses the Cosine Similarity and Hierarchical Agglomerative Cluster (HAC) analyses as the exploratory tools (Levshina 2014; Levshina 2015).

3.3 Exploring VSM with Cosine Similarity and Hierarchical Agglomerative Cluster (HAC) analysis

In VSM, cosine similarity is the popular measure for computing pairwise (dis)similarity between the target words. Cosine similarity computes the cosine of angles between the words’ vectors to capture their (dis)similarity. The cosine value between a pair of word is close to 1 when they are semantically more similar, and close to 0 when otherwise (see @ref(tab:vsm-xmpl-cossim)) (Levshina 2015, Ch. 16).

The following codes perform cosine similarity using the cossim() function from the Rling package (Levshina 2015:329).

# Cosine Similarity
mtx_cossim <- Rling::cossim(mtx_ppmi); rm(mtx_ppmi)

# Generate CosSim table output
attr(mtx_cossim, "dimnames")[[1]] <- paste("*", 
                                           attr(mtx_cossim, "dimnames")[[1]],
                                           "*",
                                           sep = "")
attr(mtx_cossim, "dimnames")[[2]] <- paste("*", 
                                           attr(mtx_cossim, "dimnames")[[2]],
                                           "*",
                                           sep = "")
knitr::kable(round(mtx_cossim, digits = 2), 
             caption = "Cosine similarity matrix between the deverbal nouns")

	bacaan	lukisan	makanan	masakan	minuman	tulisan
bacaan	1.00	0.04	0.02	0.03	0.02	0.06
lukisan	0.04	1.00	0.01	0.03	0.04	0.04
makanan	0.02	0.01	1.00	0.03	0.03	0.00
masakan	0.03	0.03	0.03	1.00	0.04	0.02
minuman	0.02	0.04	0.03	0.04	1.00	0.02
tulisan	0.06	0.04	0.00	0.02	0.02	1.00

For the cluster analysis, such as HAC (Levshina 2014; see also Gries 2013:336; Levshina 2015, Ch. 15; and Desagulier 2017:276 for R implementations on HAC), the similarity matrix/table as above needs to be converted into distance matrix as input for the cluster analysis (see the codes below adapted from Levshina 2015:330).

# Distance matrix computation
mtx_dist <- 1 - (mtx_cossim/max(mtx_cossim[mtx_cossim != 1]))
mtx_dist <- as.dist(mtx_dist)

# Cluster analysis
mtx_hcl <- hclust(mtx_dist, method = "ward.D2")

The output of HAC can be visualised into a dendrogram tree (see Figure @ref(fig:vsm-xmpl-dendrogram)). To determine the optimal number of cluster solution for grouping the nouns, we used the Average Silhouette Width (ASW) scores and tested two up to five-cluster solutions. The two-cluster solution produces the highest ASW score. See the codes below.

# ASW calculation
mtx_asw <- sapply(2:(nrow(mtx_cossim) - 1), function(x) summary(cluster::silhouette(stats::cutree(mtx_hcl, k = x), mtx_dist))$avg.width)

# names the ASW scores representing the tested cluster-solutions 
names(mtx_asw) <- 2:(nrow(mtx_cossim) - 1)

# identify the maximum ASW score
max_asw <- as.numeric(names(mtx_asw[mtx_asw == max(mtx_asw)]))

# plotting dendrogram
mtx_hcl$labels <- gsub('\\*', '', mtx_hcl$labels)
plot(mtx_hcl, hang = -1)
rect.hclust(mtx_hcl, k = max_asw)

# generate texts annotation within the plot
added_info <- paste("The ", max_asw, "-cluster solution is based on the highest Average Silhouette Width (ASW) score of ", round(max(mtx_asw), 3), ".\n(ASW ranges from 0 to 1; min. ASW score for assuming substantial cluster is 0.2)", sep = "")
mtext(added_info, side = 1, line = 1.75, cex = .8)

The following codes produce the plot visualising the number of cluster-solutions tested and their corresponding ASW scores.

plot(x = 2:5, y = mtx_asw, 
     type = "b", xaxt = "n", 
     ylab = "Average Silhouette Widths (ASW)", 
     ylim = c(0, 1),
     xlab = "Number of tested cluster solutions")
axis(side = 1, at = 2:5)

text(2:5, 
     mtx_asw, 
     labels = round(mtx_asw, digits = 3), 
     pos = 3, 
     cex = .8)
text(x = 3.25,
     y = 0.2, 
     labels = "The 2-cluster solution produces the highest ASW score")

4 Methodology

4.1 Data source

The paper uses thirteen corpus files from the Indonesian Leipzig Corpora to be trained into a vector space models. The following codes show how to load the data for the information of corpus size (i.e. "wordcount_leipzig_allcorpus.RData") and to print them.

# load the corpus size data
load("data/wordcount_leipzig_allcorpus.RData")

# print the table
leipzig.word.count %>%
  select(-Size) %>%
  rename(`Size (in word-tokens)`= Size_print,
         `Corpus files` = Corpus) %>%
  as.data.frame() %>%
  knitr::kable(row.names = TRUE, caption = "Indonesian Leipzig Corpora and their sizes")

	Corpus files	Size (in word-tokens)
1	ind_mixed_2012_1M-sentences.txt	15,052,159
2	ind_news_2008_300K-sentences.txt	5,875,376
3	ind_news_2009_300K-sentences.txt	5,868,276
4	ind_news_2010_300K-sentences.txt	5,874,158
5	ind_news_2011_300K-sentences.txt	5,852,211
6	ind_news_2012_300K-sentences.txt	5,873,523
7	ind_newscrawl_2011_1M-sentences.txt	16,376,426
8	ind_newscrawl_2012_1M-sentences.txt	16,916,778
9	ind_web_2011_300K-sentences.txt	4,472,885
10	ind_web_2012_1M-sentences.txt	15,844,629
11	ind_wikipedia_2016_1M-sentences.txt	16,506,714
12	ind-id_web_2013_1M-sentences.txt	16,406,671
13	ind-id_web_2015_3M-sentences.txt	49,849,398

In total, the thirteen corpus files amount to 180,769,204 million word-tokens.

4.2 The studied denominal verbs

The database for the studied verbs is available in the .rds, .csv, and .txt tab-delimited files named "me_parsed_nountaggedbase". The verbs have been parsed and tagged using MorphInd (Larasati, Kuboň & Zeman 2011). We limited the study on the verbs with noun-tagged roots that occur over 20 tokens. For each root, all the verb forms must occur in the three morphological schemas, namely meN-, meN-/-kan, and meN-/-i. For instance, we take mendasar ‘to be basic’, mendasari ‘to underlie sth.’, and mendasarkan ‘to base sth. on’ that are all derived from the nominal root dasar ‘base; foundation’.

The following codes show the filtering processes to retrieve the relevant verbs and Table @ref(tab:studied-verbs-retrieval) shows the snippet of the database.

# load the database
parsed_me_noun <- readRDS("data/me_parsed_nountaggedbase.rds")

# filtering parameters
min_freq <- 20 # freq. threshold
base_type <- "noun" # root category

# filtering process
df_noun <- parsed_me_noun %>% 
  filter(n > min_freq) %>% # retrieve only verbs occurring over 20 tokens
  group_by(base) %>% 
  mutate(n_affix_new = n_distinct(affix)) %>% 
  filter(n_affix_new == 3) %>% # make sure each root occurs in the 3 schemas
  ungroup()

# retrieve the target-word character vectors
me_words <- df_noun$word 


# print the database snippet
df_noun %>%
  select(-n_affix_new) %>% 
  filter(base %in% c("dasar")) %>% 
  rename(base_pos = base.pos,
         token_freq = n) %>% 
  mutate(word = paste("*", word, "*", sep = ""),
         base = paste("*", base, "*", sep = ""),
         affix = paste("*", affix, "*", sep = ""),
         morphind = gsub("(<|>)", "\\\\\\1", morphind, perl = TRUE)) %>% 
  knitr::kable(caption = "Snippet of the analysed denominal verbs")

word	token_freq	base	base_pos	morphind	affix
mendasar	4571	dasar	n	meN+dasar<n>_VSA	me
mendasari	1365	dasar	n	meN+dasar<n>+i_VSA	me.i
mendasarkan	781	dasar	n	meN+dasar<n>+kan_VSA	me.kan

In total, we analysed 51 denominal verbs based on 17 root types occurring in three morphological schemas.

We mention in the paper about checking the existence of the studied verbs with MALINDO Morph, a morphological dictionary for Indonesian and Malay (Nomoto et al. 2018). We checked it with the latest version of MALINDO Morph from the file named "malindo_dic_20181125.tsv" that has been saved as "malindo_dbase.rds" (an R type of data) and tab-delimited .csv file ("malindo_dbase.csv"). The following codes document the checking processes. The printed verbs are those absent from MALINDO Morph but available in our corpus occurring over 20 tokens overall.

# load MALINDO Dictionary data
malindo <- readRDS("data/malindo_dbase.rds")

# print the verbs that are available in the dataset but absent in MALINDO
(absent_in_malindo <- setdiff(me_words, malindo$word_form))

[1] "mengakhir"  "membuah"    "mengantung" "mewakil"   
[5] "mewaris"   

As MALINDO Morph is also based on the Leipzig Corpora, it only takes into account words occurring over ten tokens in all the 300K-sentences version of the corpus (Nomoto et al. 2018). Our frequency check of these absent verbs confirms that they all occur less than ten tokens in the 300K-sentence files that we use. The following codes show these information.

# load the freqlist per corpus and 
# check freq. of words absent in MALINDO 300K corpus
df_all_pref <- read_tsv("data/wordlist_leipzig_ME_DI_TER_percorpus.tsv", progress = FALSE)
df_all_pref %>% 
  filter(word %in% absent_in_malindo,
         grepl("300K", corpus, perl = TRUE)) %>% 
  mutate(word = paste("*", word, "*", sep = "")) %>% 
  arrange(word, corpus) %>% 
  knitr::kable(row.names = TRUE, caption = 'Token frequency of the absent verbs in the 300K-sentence files')

	corpus	word	n
1	ind_news_2008_300K	membuah	5
2	ind_news_2009_300K	membuah	3
3	ind_news_2010_300K	membuah	1
4	ind_news_2011_300K	membuah	2
5	ind_news_2012_300K	membuah	5
6	ind_web_2011_300K	membuah	1
7	ind_news_2008_300K	mengakhir	2
8	ind_news_2009_300K	mengakhir	5
9	ind_news_2010_300K	mengakhir	3
10	ind_news_2011_300K	mengakhir	6
11	ind_news_2011_300K	mengantung	1
12	ind_news_2012_300K	mengantung	1
13	ind_news_2009_300K	mewakil	2
14	ind_news_2010_300K	mewakil	2
15	ind_news_2011_300K	mewakil	4
16	ind_web_2011_300K	mewaris	1

4.3 Load the vector space models for the Indonesian Leipzig Corpora

Detailed information concerning the training parameters is available in our paper. In short, we trained the Leipzig Corpora on the MonARCH HPC using the skip-gram learning algorithm from the word2vec model (Mikolov, Chen, et al. 2013; Mikolov, Sutskever, et al. 2013; Mikolov, Yih & Zweig 2013) via the wordVectors R package (Schmidt & Li 2017). The output model is available as .bin file named "leipzig_w2v_vector_full.bin". The following codes show how the model is loaded into R using read.binary.vectors() from wordVectors package.

vsm <- read.binary.vectors("data/leipzig_w2v_vector_full.bin")

To retrieve the vector space model (VSM) of the target denominal verbs, use the following codes.

# get the VSM for the studied denominal verbs
vsm_tgt <- vsm[rownames(vsm) %in% me_words, ] 

# print the subset of the model
vsm_tgt 

A VectorSpaceModel object of  51  words and  100  vectors
                    [,1]        [,2]        [,3]        [,4]
mengatakan   -0.15838896  0.05143370  0.02442524  0.24287735
menggunakan  -0.10439277  0.07838520  0.18227956  0.38019755
mewakili     -0.25855595 -0.09868021  0.05797758  0.19394790
menempatkan   0.01809703 -0.01772323  0.24649994  0.12339904
menempati    -0.17115426 -0.10566988  0.16301627  0.17471313
mengakhiri    0.24585645  0.12317820 -0.12496730  0.18348111
melangkah     0.14984114  0.02910020  0.04853236 -0.34504056
mencontohkan -0.24159168  0.01203183  0.07734830  0.13243185
mendasar     -0.45044890 -0.07471947  0.37076223 -0.02369694
menandai      0.01873370  0.19389461  0.17055722  0.15514162
                     [,5]         [,6]
mengatakan   -0.037270460 -0.128047243
menggunakan  -0.012107351 -0.019670846
mewakili     -0.219191238 -0.049940147
menempatkan   0.090888657 -0.063201532
menempati    -0.105571389 -0.090712421
mengakhiri   -0.032905310 -0.200849608
melangkah    -0.339647442 -0.008438474
mencontohkan  0.002526026  0.068748802
mendasar      0.357669443  0.234728336
menandai      0.026616529 -0.258213133
attr(,".cache")
<environment: 0x7fbe56512620>

5 Results and discussion

The following codes run the Hierarchical Agglomerative Analysis (HAC) on the target verbs vector space model as well as the Average Silhouette Width (ASW) statistics.

# A wrapper function for HCA and ASW computations and gathering the relevant results
svs_hca <- function(vect = NULL, 
                    clust_method = c("complete", "ward.D", "ward.D2", 
                                     "single", "average", "mcquitty", 
                                     "median", "centroid")) {
  
  # 1 cosine distance using function from `wordVectors`
  cosdist <- wordVectors::cosineDist(vect, vect)
  cosdist <- as.dist(cosdist)
  
  # 2 Hierarchical Cluster Analysis
  hca <- hclust(cosdist, method = clust_method)
  
  # 3 Compute the 'average silhouette width' (ASW) 
  #   using the 'cutree' function for HCA partitioning
  asw_f <- function(x) {
    summary(cluster::silhouette(stats::cutree(hca, k = x), cosdist))$avg.width
  }
  asw_cutree <- sapply(2:(dim(vect)[1] - 1), asw_f)
  names(asw_cutree) <- 2:(dim(vect)[1] - 1)
  
  # 3.2 get the cluster number with highest ASW score
  n_cluster <- as.numeric(names(asw_cutree[asw_cutree == max(asw_cutree)]))
  max_asw <- asw_cutree[asw_cutree == max(asw_cutree)]
  
  # 4 Put the results into a list
  res <- list(cosdist, hca, n_cluster, max_asw, asw_cutree)
  names(res) <- c("cosine_dist", "hcluster", "n_cluster", "asw", "asw_all")
  return(res)
}

# Hierarchical Agglomerative Cluster (HAC) analysis for the denominal verbs
clust_method <- "ward.D2"
hca_res <- svs_hca(vect = vsm_tgt,
                   clust_method = clust_method)

The following codes generate the plot for the ASW scores in the paper.

# preparing data frame for the ASW and tested clusters
asw_df <- hca_res$asw_all
asw_df <- tibble::tibble(tested_cluster = names(asw_df),
                         asw = unname(asw_df))
asw_df_sort <- dplyr::arrange(asw_df, dplyr::desc(asw))

# plot with ggplot2
asw_df_sort %>% 
  ggplot(aes(x = reorder(tested_cluster, asw), y = asw, group = 1)) +
  geom_step() +
  ylim(c(0, 1)) +
  coord_flip() +
  theme_light() +
  geom_text(aes(label = round(asw, 3)), hjust = -0.2, size = 2.5) +
  labs(x = "Number of tested cluster solutions",
       y = "Average Silhouette Width (ASW)",
       caption = paste("The tested cluster solution ranges from 2 to ", 
                       length(me_words) - 1, 
                       " (i.e., the length of the analysed words (", 
                       length(me_words), ") - 1).\nThe highest ASW of ", 
                       round(hca_res$asw, 3), 
                       " is for ", hca_res$n_cluster, 
                       "-cluster solution", 
                       sep = ""))

The dendrogram tree is generated using the following codes:

opar <- par(no.readonly = TRUE)
par(mar = c(2, 2, 2, 10))
dend <- as.dendrogram(hca_res$hcluster)
dend <- dendextend::set(dend = dend, "labels_cex", .9)
plot(dend, horiz = TRUE)
dendextend::rect.dendrogram(dend,
                            k = hca_res$n_cluster,
                            horiz = TRUE,
                            border = 2,
                            lwd = 0.35)

par(opar)

# A function to retrieve the token frequency of the studied verbs
wfreq <- function(df = df_noun, w, formatted = TRUE) {
  freq <- dplyr::pull(dplyr::filter(df, word == w), n)
  if (formatted) {
    freq <- format(freq, big.mark = ",")
  } 
  return(freq)
}

5.1 Semantic cluster

The subcluster of the motion verbs is extracted using the following codes:

clusters <- dendextend::cutree(hca_res$hcluster, k = hca_res$n_cluster)
opar <- par(no.readonly = TRUE)
par(mar = c(2, 2, 2, 7))
dend <- as.dendrogram(hca_res$hcluster)
to_prune <- names(clusters[!names(clusters) %in% c("melangkahkan", "melangkah", "menapak", "menapaki", "menapakkan", "menjejakkan", "menjejak", "menjejaki")])
plot(dendextend::prune(dend, to_prune), horiz = TRUE)
par(opar)

In this section, we present three tables for n-grams data of verbs with the root tapak ‘sole of the foot’. The data can be loaded as follows:

tapak <- readr::read_tsv("data/ngramexampl_3gr_menapak.txt")
tapaki <- readr::read_tsv("data/ngramexampl_3gr_menapaki.txt")
tapakkan <- readr::read_tsv("data/ngramexampl_3gr_menapakkan.txt")

The following codes generate the ten most frequent right-side 3-gram for menapak.

tapakgr <- tapak %>%
  filter(w1 == "menapak") %>% 
  count(ngrams, sort = TRUE) %>% 
  .[1:10, ] %>% 
  mutate(ngrams = paste("*", ngrams, "*", sep = "")) # make ngrams italics
knitr::kable(tapakgr, caption = "The ten most frequent 3-gram for *menapak*", row.names = TRUE)

	ngrams	n
1	menapak_masa_depan	10
2	menapak_ke_babak	6
3	menapak_di_jalan	4
4	menapak_di_lantai	4
5	menapak_karir_di	4
6	menapak_di_atas	3
7	menapak_tilas_jejak	3
8	menapak_di_bumi	2
9	menapak_di_jalanan	2
10	menapak_di_permukaan	2

Menapak can be used as transitive (item 1, 5, and 7) and intransitive verbs (the remaining items in Table @ref(tab:tapak-gram-table)). Its transitive usage shares similar right-side collocation patterns with the MeN-/-i form menapaki, especially their direct object with masa depan ‘future’ and karir ‘career’ (Table@ref(tab:tapaki-gram-table) below).

tapakigr <- tapaki %>%
  filter(w1 == "menapaki") %>% 
  count(ngrams, sort = TRUE) %>% 
  .[1:10, ] %>% 
  mutate(ngrams = paste("*", ngrams, "*", sep = ""))
knitr::kable(tapakigr, caption = "The ten most frequent 3-gram for *menapaki*", row.names = TRUE)

	ngrams	n
1	menapaki_anak_tangga	11
2	menapaki_jalan_menuju	11
3	menapaki_masa_depan	11
4	menapaki_jalan_yang	9
5	menapaki_karir_di	9
6	menapaki_karier_sebagai	7
7	menapaki_karier_di	6
8	menapaki_babak_baru	5
9	menapaki_dunia_kerja	5
10	menapaki_karir_sebagai	5

This is different from the transitive usage with MeN-/-kan schema (Table @ref(tab:tapakkan-gram-table) below), which predominantly has kaki ‘foot’ as its direct object, followed by either locational/directional prepositional phrses or motion verb complements (e.g., memasuki ‘to enter’ [item 5] and maju ‘to move forward’ [item 8]).

tapakkangr <- tapakkan %>%
  filter(w1 == "menapakkan") %>% 
  count(ngrams, sort = TRUE) %>% 
  .[1:10, ] %>% 
  mutate(ngrams = paste("*", ngrams, "*", sep = ""))
knitr::kable(tapakkangr, caption = "The ten most frequent 3-gram for *menapakkan*", row.names = TRUE)

	ngrams	n
1	menapakkan_kakinya_di	24
2	menapakkan_kaki_di	14
3	menapakkan_kaki_ke	3
4	menapakkan_dirinya_di	2
5	menapakkan_kaki_memasuki	2
6	menapakkan_kaki_saat	2
7	menapakkan_kakinya_ke	2
8	menapakkan_kakinya_maju	2
9	menapakkan_bisnis_toko	1
10	menapakkan_citra_donnie	1

Codes for extracting the subset of the mixture of communication and psych verbs are as follows.

clusters <- dendextend::cutree(hca_res$hcluster, k = hca_res$n_cluster)
opar <- par(no.readonly = TRUE)
par(mar = c(2, 2, 2, 7))
dend <- as.dendrogram(hca_res$hcluster)
to_prune <- names(clusters[!names(clusters) %in% c("membayangkan", "menyesal", "menyesali", "mengatai")])
plot(dendextend::prune(dend, to_prune), horiz = TRUE)
par(opar)

The above dendrogram is from the top cluster in Figure @ref(fig:hca-plot), consisting of (i) menyesal ‘to be regretful’ (N = 2,556), (ii) menyesali ‘regret sth.’ (N = 956), and (iii) membayangkan ‘imagine; visualise’ (N = 2,719). The other subset of communication and psych verbs (i.e., mengatakan ‘to say sth.’ (N = 265,381), mencontohkan ‘to exemplify’ (N = 4,799), and menyesalkan ‘regret sth.’ (N = 1,976)) is extracted as follows.

clusters <- dendextend::cutree(hca_res$hcluster, k = hca_res$n_cluster)
opar <- par(no.readonly = TRUE)
par(mar = c(2, 2, 2, 7))
dend <- as.dendrogram(hca_res$hcluster)
to_prune <- names(clusters[!names(clusters) %in% c("mencontohkan", "mengatakan", "menyesalkan")])
plot(dendextend::prune(dend, to_prune), horiz = TRUE)
par(opar)

What interesting between verbs in these last two clusters is that the MeN-/-i (i.e. mengatai) and MeN-/-kan verbs (i.e. mengatakan) with the root kata ‘word’ are way apart in the dendrogram. Similar case is apparent between menyesalkan separated with menyesal and menyesali, where the latter two verbs cluster together and are merged first in Figure @ref(fig:subcluster-psychs-1) (see §@ref(cluster-split) for further discussion on this split).

5.2 Root-based clustering

This cluster type captures denominal verbs of a given root with the three different morphological schemas that cluster together (Figure @ref(fig:subcluster-root-based)). We have seen few examples of these in the motion cluster with the root tapak ‘sole of the foot’ and jejak ‘footprint’, the derived forms of which fall into one cluster but differ in terms of their within-cluster branching (Figure @ref(fig:subclust-motion)). The other examples are based on the following roots:

1. susu ‘milk’ -> menyusu ‘(of a baby/young animal) to suckle’ (N = 458); menyusui ‘to breast-feed sb.’ (N = 2,538); menyusukan ‘to let sb. suckle; to breast-feed sb.’ (N = 35)
2. dasar ‘base’ -> mendasar ‘basic; foundational’ (N = 4,571); mendasari ‘to underlie sth.’ (N = 1,365); mendasarkan ‘to base X (on Y)’ (N = 781)
3. tempat ‘place; location’ -> menempat ‘to place/position (sth.) at’ (N = 27); menempati ‘to occupy; to site at’ (N = 11,150); menempatkan ‘to put/place/position X at Y’ (N = 11,513)
4. wakil ‘vice; representative; deputy’ -> mewakil ‘to (be a) represent(ative of) X’ (N = 39); mewakili ‘to (be a) represent(ative of) X’ (N = 12,389); mewakilkan ‘to assign sb. as a representative’ (N = 220)

The cluster subsets of these verbs are extracted from Figure @ref(fig:hca-plot) into Figure @ref(fig:subcluster-root-based) with the codes below.

clusters <- dendextend::cutree(hca_res$hcluster, k = hca_res$n_cluster)
opar <- par(no.readonly = TRUE)
par(mar = c(2, 2, 2, 7))
dend <- as.dendrogram(hca_res$hcluster)
to_prune <- names(clusters[!names(clusters) %in% c("mewakili", "mewakilkan", "mewakil", "menyusu", "menyusui", "menyusukan", "menempati", "menempatkan", "menempat", "mendasari", "mendasarkan", "mendasar")])
plot(dendextend::prune(dend, to_prune), horiz = TRUE)
dendextend::rect.dendrogram(dendextend::prune(dend, to_prune),
                            k = 4,
                            horiz = TRUE,
                            border = 2,
                            lwd = 0.35)

par(opar)

In the paper, we present n-grams data contrasting mewakili and mewakilkan in terms of their ride-side collocates in their 3-grams data. First, the codes for generating n-grams for mewakili ‘to (be a) represent(ative of) X’ are shown below.

mewakili <- readr::read_tsv("data/ngramexampl_5gr_mewakili.txt", progress = FALSE)
mewakili_mini <- dplyr::filter(mewakili, w1 == "mewakili")
mewakili_count <- dplyr::count(mewakili_mini, w1, w2, w3, sort = TRUE)
mewakili_print <- mewakili_count[1:10,]
mewakili_print <- dplyr::mutate(mewakili_print, 
                                ngrams = paste(w1, "_", w2, "_", w3, sep = ""),
                                ngrams = paste("*", ngrams, "*", sep = ""))
mewakili_print <- dplyr::select(mewakili_print, ngrams, n)
knitr::kable(mewakili_print, caption = "10 most frequent 3-gram for *mewakili* 'to (be a) represent(ative of) X'", row.names = TRUE)

	ngrams	n
1	mewakili_kebijakan_editorial	174
2	mewakili_indonesia_di	142
3	mewakili_indonesia_dalam	129
4	mewakili_indonesia_pada	69
5	mewakili_lebih_dari	38
6	mewakili_iklan_anda	36
7	mewakili_kepala_dinas	35
8	mewakili_indonesia_untuk	32
9	mewakili_indonesia_ke	26
10	mewakili_kepala_badan	23

Then, the codes for 3-grams of mewakilkan ‘to make X as the representative (of Y)’.

mewakilkan <- readr::read_tsv("data/ngramexampl_5gr_mewakilkan.txt", progress = FALSE)
mewakilkan_mini <- dplyr::filter(mewakilkan, w1 == "mewakilkan")
mewakilkan_count <- dplyr::count(mewakilkan_mini, w1, w2, w3, sort = TRUE)
mewakilkan_print <- mewakilkan_count[1:10,]
mewakilkan_print <- dplyr::mutate(mewakilkan_print, 
                                ngrams = paste(w1, "_", w2, "_", w3, sep = ""),
                                ngrams = paste("*", ngrams, "*", sep = ""))
mewakilkan_print <- dplyr::select(mewakilkan_print, ngrams, n)
knitr::kable(mewakilkan_print, caption = "10 most frequent 3-gram for *mewakilkan* 'to appoint/select X as the representative of Y'", row.names = TRUE)

	ngrams	n
1	mewakilkan_sebuah_film	6
2	mewakilkan_kepada_orang	5
3	mewakilkan_orang_lain	4
4	mewakilkan_benua_asia	3
5	mewakilkan_kehadirannya_kepada	3
6	mewakilkan_kepada_unais	3
7	mewakilkan_kepada_wakil	3
8	mewakilkan_6_perwakilan	2
9	mewakilkan_bisa_dengan	2
10	mewakilkan_dirinya_lewat	2

5.3 Clustering-split between morphological schemas for a given root

This sub-section addresses in more details split cases between morphological schemas for a given root. The split, especially between MeN-/-kan and MeN-/-i verbs, reflects Sneddon et al’s (2010:100–101) hypothesis concerning clear semantic difference between some of a pair of MeN-/-kan and MeN-/-i verbs with the same root. Our VSM-based approach allows us to visualise such split through the dendrogram based on large-scale usage data. The sub-section also demonstrates further enrichment in charactersing the difference between morphologically related verb pairs using the technique of nearest neighbours based on the VSM data.

In §@ref(semclust), we have mentioned the clear split between the transitive melangkahkan ‘to move the foot forward’ and melangkahi ‘to step over’. Looking at the 2-gram data for each verb shows that they have different semantic orientation. The following codes generate the 2-gram data for melangkahi.

langkahi <- readr::read_tsv("data/ngramexampl_3gr_melangkahi.txt")
langkahi_print <- langkahi %>% 
  filter(w1 == "melangkahi") %>% 
  mutate(ngrams = paste(w1, "_", w2, sep = ""),
         ngrams = paste("*", ngrams, "*", sep = "")) %>% 
  count(ngrams, sort = TRUE) %>% 
  .[1:10, ]
knitr::kable(langkahi_print, caption = "Ten most frequent 2-gram for right-side collocates of *melangkahi*", row.names = TRUE)

	ngrams	n
1	melangkahi_kewenangan	7
2	melangkahi_aturan	6
3	melangkahi_batas-batas	4
4	melangkahi_apa	3
5	melangkahi_beberapa	3
6	melangkahi_mekanisme	3
7	melangkahi_pundak	3
8	melangkahi_tlundak	3
9	melangkahi_batasan	2
10	melangkahi_dasar-dasar	2

Melangkahi predominantly conveys metaphorical sense related to disobeying/disregarding certain (i) rules/protocols (i.e., aturan, batas-batas/batasan ‘limits; restriction’, mekanisme ‘mechanism’), (ii) foundation (dasar-dasar), or (iii) authority (kewenangan).

In contrast, melangkahkan predominantly collocates with kaki ‘foot’ as its direct object collocates, which can be used in the literal, translational motion and metaphorical motion (see our paper for the example sentences).

langkahkan <- readr::read_tsv("data/ngramexampl_3gr_melangkahkan.txt")
langkahkan_print <- langkahkan %>% 
  filter(w1 == "melangkahkan") %>% 
  mutate(ngrams = paste(w1, "_", w2, sep = ""),
         ngrams = paste("*", ngrams, "*", sep = "")) %>% 
  count(ngrams, sort = TRUE) %>% 
  .[1:10, ]
knitr::kable(langkahkan_print, caption = "Ten most frequent 2-gram for right-side collocates of *melangkahkan*", row.names = TRUE)

	ngrams	n
1	melangkahkan_kaki	162
2	melangkahkan_kakinya	123
3	melangkahkan_kakiku	13
4	melangkahkan_kedua	3
5	melangkahkan_satu	3
6	melangkahkan_kakimu	2
7	melangkahkan_bakti	1
8	melangkahkan_bidak	1
9	melangkahkan_gambaran	1
10	melangkahkan_jalan	1

Observation on the n-gram can be enriched using information from the VSM of words. Given that the skip-gram algorithm of word2vec learns to predict the contextual environments given a target word (cf. Mikolov, Chen, et al. 2013), one can retrieve from the VSM a set of words that have similar contextual-vector distribution to a given target verb on the basis their cosine similarities; these words can be metaphorically referred to as the verb’s nearest neighbours. Table @ref(tab:nearest-to-melangkahi) illustrates the idea for melangkahi ‘to step over’.

# function to retrieve and print top-10 nearest neighbours
neighbours <- function(model, seed_word, n = 10) {
  topn <- n + 1
  df <- wordVectors::closest_to(model, seed_word, topn)
  
  # remove the seed words because it is similar to itself and become the first nearest word. We thus count top-10 nearest neighbours from the rank 2 to 11
  df <- dplyr::filter(df, word != seed_word) 
  df <- dplyr::mutate(df, word = paste("*", word, "*", sep = ""))
  df <- as.data.frame(df)
  colnames(df)[2] <- paste('similarity to "*', seed_word, '*"', sep = "")
  return(df)
}

# get the nearest neighbours for *melangkahi*
near_langkahi <- neighbours(vsm, seed_word = "melangkahi")
knitr::kable(near_langkahi, 
             caption = "10 closest words to *melangkahi* 'to step over'",
             row.names = TRUE)

	word	similarity to “melangkahi”
1	mengangkangi	0.5508479
2	berkeras	0.5435145
3	memperhitungkannya	0.5337096
4	mengacuhkan	0.5174865
5	memagari	0.5163426
6	memegang	0.5040103
7	membelakangi	0.5036757
8	mematuhi	0.4959915
9	bersikeras	0.4949589
10	berbenturan	0.4929108

The closest words may not necessarily similar in meaning (e.g., near-synonyms), but may exhibit different kind of relationships, such as antonyms. Words in Table @ref(tab:nearest-to-melangkahi) conveying more or less antonymous sense to melangkahi ‘to step over; to disregard’ include mengacuhkan ‘to care about/heed sth.’, mematuhi ‘to obey’, and (to a degree) memperhitungkannya ‘to take sth. into account’. Mengangkangi ‘to straddle sth.’ is the closest in meaning with melangkahi as it can be extended into ‘disregarding’ sense from its physical, posture sense: the 2-gram data for mengangkangi across the whole corpus (Table @ref(tab:kangkangi-gr)) reveals that it does co-occur with rules-related direct objects, such as hukum ‘law’ (3 tokens), peraturan ‘regulation’ (3), kebenaran ‘the truth’ (2), prinsip ‘principles’ (2), undang-undang ‘constitution’ (2), aturan ‘rules’ (1), inter alia.

ngangkangi <- readr::read_tsv("data/ngramexampl_3gr_mengangkangi.txt")
ngangkangi_print <- ngangkangi %>% 
  filter(w1 == "mengangkangi") %>% 
  mutate(ngrams = paste(w1, "_", w2, sep = ""),
         ngrams = paste("*", ngrams, "*", sep = "")) %>% 
  count(ngrams, sort = TRUE)
knitr::kable(ngangkangi_print[ngangkangi_print$n > 1, ], caption = "The 2-gram data for right-side collocates of *mengangkangi* (n >= 2)", row.names = TRUE)

	ngrams	n
1	mengangkangi_tubuh	4
2	mengangkangi_dunia	3
3	mengangkangi_hukum	3
4	mengangkangi_peraturan	3
5	mengangkangi_seluruh	3
6	mengangkangi_bagian	2
7	mengangkangi_hak	2
8	mengangkangi_jembatan	2
9	mengangkangi_kebenaran	2
10	mengangkangi_mu	2
11	mengangkangi_prinsip	2
12	mengangkangi_sebuah	2
13	mengangkangi_tanah	2
14	mengangkangi_undang-undang	2

The following codes retrieve words nearest to melangkahkan.

knitr::kable(neighbours(vsm, "melangkahkan", n = 10), caption = "Ten closest words to *melangkahkan*", row.names = TRUE)

	word	similarity to “melangkahkan”
1	menjejakkan	0.7487976
2	melangkah	0.7351338
3	dilangkahkan	0.7294263
4	berlari	0.7258124
5	kakiku	0.7195594
6	menghunjamkan	0.7150664
7	menapakkan	0.7126413
8	berjingkat	0.7079603
9	kakinya	0.7068803
10	langkahkan	0.7066612

The codes to retrieve the nearest words of mengatai ‘to rebuke; speake of one’s badness’.

knitr::kable(neighbours(vsm, "mengatai", n = 10), caption = "Ten closest words to *mengatai*", row.names = TRUE)

	word	similarity to “mengatai”
1	memaki	0.7476359
2	marah-marah	0.6818458
3	cerewet	0.6806308
4	mengejek	0.6795740
5	memaki-maki	0.6713904
6	jengkel	0.6675079
7	diejek	0.6645705
8	diolok-olok	0.6641099
9	meledek	0.6628723
10	berbohong	0.6597248

In contrast, mengatakan mostly appears as communication verb with similar distribution with other reported speech verbs.

knitr::kable(neighbours(vsm, "mengatakan", n = 10), caption = "Ten closest words to *mengatakan*", row.names = TRUE)

	word	similarity to “mengatakan”
1	menegaskan	0.8376837
2	menyatakan	0.8318030
3	mengungkapkan	0.8079668
4	mengemukakan	0.7967164
5	menuturkan	0.7925858
6	menjelaskan	0.7808896
7	menyebutkan	0.7640667
8	menerangkan	0.7568715
9	mengakui	0.7495804
10	mengatkan	0.7467988

Nearest words to membuahi ‘to breed sth.’ are retrieved as follows.

knitr::kable(neighbours(vsm, "membuahi", n = 10), caption = "Ten closest words to *membuahi*", row.names = TRUE)

	word	similarity to “membuahi”
1	dibuahi	0.8372309
2	ovum	0.8330801
3	sperma	0.7965994
4	gamet	0.7326975
5	pembuahan	0.7300101
6	terbuahi	0.7016097
7	spermatozoid	0.6901719
8	spermatozoa	0.6794663
9	parthenogenesis	0.6775866
10	zigot	0.6712285

Codes for extracting closest words to membuahkan ‘to bear a fruit; to result in sth.’.

knitr::kable(neighbours(vsm, "membuahkan", n = 10), caption = "Ten closest words to *membuahkan*", row.names = TRUE)

	word	similarity to “membuahkan”
1	berbuah	0.6716316
2	mem-buahkan	0.6296921
3	tercipta	0.6214626
4	membuah	0.5991543
5	menuai	0.5729809
6	tendangannya	0.5533927
7	ditepis	0.5530693
8	kerasnya	0.5528045
9	pinalti	0.5482891
10	dimentahkan	0.5478375

Notes on the usage sentences for membuah are below. But first load the sentence citations for membuah into R and print them into the console. After that the notes are manually created by manually inspecting all the usage sentences. The sentence data is available as .rds file of list (sentence_membuah.rds) and .txt file as plain text of sentences (sentence_membuah.txt).

membuah <- readRDS("data/sentence_membuah.rds")
membuah$ind_mixed_2012_1M # for instance, retrieve sentences found in `ind_mixed_2012_1M-sentences.txt`

[1] "ind_mixed_2012_1M__314113__Dan , entah sudah berapa kali pejabat di Dinas Tata Kota dan Permukiman ( DTKP ) Pemkot telah berganti , tetapi persoalan bangunan mangkrak itu tak juga  <m>membuah</m>  solusi cepat ."
[2] "ind_mixed_2012_1M__635047__Pertemuan dengan ibu saudaranya  <m>membuah</m>  seribu kegembiraan ."                                                                                                                   
[3] "ind_mixed_2012_1M__912835__Peluang Irak lewat tendangan Younis Khalef di dalam kotak penalti juga masih belum  <m>membuah</m>  hasil ."                                                                             
[4] "ind_mixed_2012_1M__912841__Namun penyerbuan ke al-Karak , 1183-84 tidak  <m>membuah</m>  hasil yang memuaskan ."                                                                                                    

The sentence format is corpus-file-name_sentence-id-number_sentence-citation. The <m>...</m> tag indicates the corresponding match/word/verb.

• membuah (56 citations; Summary: analogy to membuahkan = 24 citations; split with -kan = 1; mispell for membuat = 25; mispell for membuang = 5; unclear = 1)

  • ind_mixed_2012_1M: 4 citations (all FULL FORM)
  • ind_news_2008_300K: 5 citations (2 mispell for membuat [id 53666 & 155569]; 3 FULL FORM analogy to membuahkan)
  • ind_news_2009_300K: 3 citations (1 mispell for membuat [id 200027] 2 FULL FORM analogy to membuahkan)
  • ind_news_2010_300K: 1 citation (all mispell for membuat)
  • ind_news_2011_300K: 2 citations (1 mispell for membuat [id 266814]; 1 FULL FORM analogy to membuahkan)
  • ind_news_2012_300K: 5 citations (4 mispell for membuat [id 200952, 206980, 268407, 269898]; 1 FULL FORM analogy to membuahkan)
  • ind_newscrawl_2011_1M: 6 citations (3 FULL FORM analogy to membuahkan; 3 mispell for membuat [id 118419, 151095, 205239])
  • ind_newscrawl_2012_1M: 3 citations (1 mispell for membuat [id 201217]; 1 mispell for membuang [id 401158]; 1 got split with -kan [id 595662])
  • ind_web_2011_300K: 1 citation (mispell for membuat)
  • ind_web_2012_1M: 4 citations (1 mispell for membuat [id 602274]; 1 unclear for mispell for membuat [id 365510]; 2 FULL FORM analogy to membuahkan)
  • ind_wikipedia_2016_1M: 5 citations (2 mispell for membuat [id 320981, 596205]; 3 FULL FORM analogy to membuahkan)
  • ind-id_web_2013_1M: 5 citations (1 mispell for membuang [id 886076]; 4 FULL FORM as membuahkan)
  • ind-id_web_2015_3M: 12 citations (1 FULL FORM as membuahkan [id 1396285]; 3 mispell for membuang [id 1362223, 1549653, 1564536], the remaining 8 mispell for membuat)

The following codes retrieve the nearest words to mengakhiri (N = 8,512).

knitr::kable(neighbours(vsm, "mengakhiri", n = 10), caption = "Ten closest words to *mengakhiri*", row.names = TRUE)

	word	similarity to “mengakhiri”
1	mengakhir	0.7149269
2	menyudahi	0.6486199
3	mengahiri	0.6346930
4	akhiri	0.6329971
5	berakhir	0.6263411
6	mengakhirinya	0.6060989
7	mengakiri	0.5811454
8	memimpin	0.5763634
9	memperpanjang	0.5681369
10	memupus	0.5669336

If reader wishes to check all usage sentences for mengakhir to confirm that it occurs as a full word-form and has similar usage patterns with meN-/-i form mengakhiri (i.e. in transitive constructions), use the following code. It will print all sentences for mengakhir into R console.

mengakhir <- readRDS("data/sentence_mengakhir.rds")
unlist(unname(mengakhir)) # print all 57 sentences in the console

Next, the nearest words to mengakhirkan (N = 116) are shown below.

knitr::kable(neighbours(vsm, "mengakhirkan", n = 10), caption = "Ten closest words to *mengakhirkan*", row.names = TRUE)

	word	similarity to “mengakhirkan”
1	shalat	0.8595279
2	menjamak	0.8582150
3	shubuh	0.8510190
4	dijamak	0.8504550
5	zhuhur	0.8434480
6	qashar	0.8383760
7	isya	0.8300283
8	disunnahkan	0.8250431
9	menjama	0.8227094
10	qabliyah	0.8215906

5.4 Issues and challenges

Codes to extract the nearest words to mengantung ‘to hang’ (N = 28), which is a misspelling of menggantung ‘to hang’ (with double g) (N = 1,100). Nearest words output shows that mengantung is indeed the closest word to menggantung in their usage co-occurrence.

knitr::kable(neighbours(vsm, "mengantung", n = 10), caption = "Ten closest words to *mengantung*", row.names = TRUE)

	word	similarity to “mengantung”
1	menggantung	0.6580294
2	blandar	0.6541215
3	memaku	0.6125143
4	ditelungkupkan	0.5888941
5	dibaringkan	0.5863362
6	loso	0.5808274
7	digergaji	0.5766285
8	tertelungkup	0.5721740
9	tersekap	0.5694923
10	ditindih	0.5677198

The codes below extract the nearest words for mengantungkan (N = 44), a misspelling for menggantungkan ‘to hang sth. (onto sth.)’ (N = 1,264).

knitr::kable(neighbours(vsm, "mengantungkan", n = 10), caption = "Ten closest words to *mengantungkan*", row.names = TRUE)

	word	similarity to “mengantungkan”
1	menggantungkan	0.7764547
2	penghidupannya	0.6517582
3	peladang	0.5890552
4	matapencaharian	0.5649209
5	mengentaskannya	0.5516661
6	bersawah	0.5515007
7	petani-petani	0.5469401
8	pengais	0.5458413
9	bertani	0.5404030
10	upahan	0.5370267

Then, the nearest words for mengantungi (N = 327) which is closest to the common spelling with mengantongi (N = 3,574) based on the root kantong ‘pocket’.

knitr::kable(neighbours(vsm, "mengantungi", n = 10), caption = "Ten closest words to *mengantungi*", row.names = TRUE)

	word	similarity to “mengantungi”
1	mengantongi	0.7021752
2	mengoleksi	0.6954547
3	memuncaki	0.6904767
4	raihan	0.6513341
5	terpaut	0.6435489
6	mengungguli	0.6411234
7	torehan	0.6217388
8	pemuncak	0.6164198
9	mengemas	0.6040049
10	diposisi	0.6004999

Finally, the case where part of the complex word is split. The paper illustrates this with menanda which is part of menandatangani ‘to give signature; to sign’ but is written separately with whitespace, thus, menanda tangani, where the whitespace got tokenised.

knitr::kable(neighbours(vsm, "menanda", n = 10), caption = "Ten closest words to *menanda*", row.names = TRUE)

	word	similarity to “menanda”
1	ditanda	0.7982222
2	menandatangani	0.6721747
3	menanda-tangani	0.6597414
4	menandatangai	0.6295324
5	menandatangi	0.6148956
6	tanganinya	0.6112830
7	ditandatangani	0.6063122
8	ditanda-tangani	0.6057313
9	ditandatanganinya	0.5950392
10	meneken	0.5891363

• menanda (N = 121; Summary: split = 113; intransitive usage = 5; transitive usage analogy = 2; ambiguous = 1)

  • ind_mixed_2012_1M (22 citations: 18 split from menandatangani; 4 citations as intransitive with pada (138307, 278193, 542191, 754529))
  • ind_news_2008_300K (1 citation: 1 split from menandatangani)
  • ind_news_2009_300K (2 citations: 2 split from menandatangani)
  • ind_news_2010_300K (2 citations: 2 split from menandatangani)
  • ind_news_2011_300K (3 citations: 3 split from menandatangani)
  • ind_newscrawl_2011_1M (10 citations: 10 split from menandatangani)
  • ind_newscrawl_2012_1M (9 citations: 9 split from menandatangani)
  • ind_web_2011_300K (4 citations: 4 split from menandatangani)
  • ind_web_2012_1M (18 citations):
    • 14 citations split from menandatangani
    • 1 citation of transitive usage in analogy to menandai (44043): “II . mangarus rupa , menerka , mencoba <m>menanda</m> siapa orang ini .”
    • 1 citation of transitive usage in analogy to menandakan ‘to mark/signal (the existence of) sth.’ (368004): “Majlis tamat , hadir berduyun sekumpulan penari dengan meriahnya , <m>menanda</m> temasya kecil kampung bakal bermula , sedikit demi sedikit sahutan penduduk kampung menggila sepenuh nafsu menari sekali bersama penari .” (Malaysian Malay?)
    • 1 citation of intransitive usage (544395): “Bagi aku , kalau tidak sempat <m>menanda</m> di sekolah ( memang tak cukup masa ) , terpaksalah bawa di rumah .”
    • 1 citation of ambiguous usage (981384): “Nama dan No Kad Pengenalan disebut , dan mereka <m>menanda</m> .”
  • ind_wikipedia_2016_1M (16 citations: 16 split from menandatangan(i))
  • ind-id_web_2013_1M (16 citations: 16 split from menandatangan(i))
  • ind-id_web_2015_3M (18 citations: 18 split from menandatangan(i))

Session Info

devtools::session_info()

─ Session info ─────────────────────────────────────────────────

─ Packages ─────────────────────────────────────────────────────
 package     * version date       lib
 assertthat    0.2.1   2019-03-21 [1]
 backports     1.1.5   2019-10-02 [1]
 broom         0.5.5   2020-02-29 [1]
 callr         3.2.0   2019-03-15 [1]
 cellranger    1.1.0   2016-07-27 [1]
 cli           2.0.2   2020-02-28 [1]
 cluster     * 2.1.0   2019-06-19 [2]
 colorspace    1.4-1   2019-03-18 [1]
 crayon        1.3.4   2017-09-16 [1]
 DBI           1.0.0   2018-05-02 [1]
 dbplyr        1.4.2   2019-06-17 [1]
 dendextend  * 1.13.2  2019-12-02 [1]
 desc          1.2.0   2018-05-01 [1]
 devtools      2.2.1   2019-09-24 [1]
 digest        0.6.25  2020-02-23 [1]
 dplyr       * 0.8.5   2020-03-07 [1]
 ellipsis      0.3.0   2019-09-20 [1]
 evaluate      0.14    2019-05-28 [1]
 fansi         0.4.1   2020-01-08 [1]
 farver        2.0.1   2019-11-13 [1]
 forcats     * 0.5.0   2020-03-01 [1]
 fs            1.3.1   2019-05-06 [1]
 generics      0.0.2   2018-11-29 [1]
 ggplot2     * 3.3.0   2020-03-05 [1]
 glue          1.3.2   2020-03-12 [1]
 gridExtra     2.3     2017-09-09 [1]
 gtable        0.3.0   2019-03-25 [1]
 haven         2.2.0   2019-11-08 [1]
 highr         0.8     2019-03-20 [1]
 hms           0.5.3   2020-01-08 [1]
 htmltools     0.3.6   2017-04-28 [1]
 httr          1.4.1   2019-08-05 [1]
 jsonlite      1.6     2018-12-07 [1]
 knitr         1.28    2020-02-06 [1]
 labeling      0.3     2014-08-23 [1]
 lattice       0.20-38 2018-11-04 [2]
 lifecycle     0.2.0   2020-03-06 [1]
 lubridate     1.7.4   2018-04-11 [1]
 magrittr      1.5     2014-11-22 [1]
 memoise       1.1.0   2017-04-21 [1]
 modelr        0.1.5   2019-08-08 [1]
 munsell       0.5.0   2018-06-12 [1]
 nlme          3.1-144 2020-02-06 [2]
 pillar        1.4.3   2019-12-20 [1]
 pkgbuild      1.0.3   2019-03-20 [1]
 pkgconfig     2.0.3   2019-09-22 [1]
 pkgload       1.0.2   2018-10-29 [1]
 prettyunits   1.0.2   2015-07-13 [1]
 processx      3.3.1   2019-05-08 [1]
 ps            1.3.0   2018-12-21 [1]
 purrr       * 0.3.3   2019-10-18 [1]
 R6            2.4.1   2019-11-12 [1]
 Rcpp          1.0.4   2020-03-17 [1]
 readr       * 1.3.1   2018-12-21 [1]
 readxl        1.3.1   2019-03-13 [1]
 remotes       2.1.0   2019-06-24 [1]
 reprex        0.3.0   2019-05-16 [1]
 rlang         0.4.5   2020-03-01 [1]
 Rling       * 1.0     2019-09-12 [1]
 rmarkdown     2.1     2020-01-20 [1]
 rprojroot     1.3-2   2018-01-03 [1]
 rstudioapi    0.10    2019-03-19 [1]
 rvest         0.3.5   2019-11-08 [1]
 scales        1.1.0   2019-11-18 [1]
 sessioninfo   1.1.1   2018-11-05 [1]
 stringi       1.4.6   2020-02-17 [1]
 stringr     * 1.4.0   2019-02-10 [1]
 testthat      2.3.1   2019-12-01 [1]
 tibble      * 2.1.3   2019-06-06 [1]
 tidyr       * 1.0.2   2020-01-24 [1]
 tidyselect    1.0.0   2020-01-27 [1]
 tidyverse   * 1.3.0   2019-11-21 [1]
 usethis       1.5.1   2019-07-04 [1]
 vctrs         0.2.4   2020-03-10 [1]
 viridis       0.5.1   2018-03-29 [1]
 viridisLite   0.3.0   2018-02-01 [1]
 withr         2.1.2   2018-03-15 [1]
 wordVectors * 2.0     2019-06-01 [1]
 xfun          0.12    2020-01-13 [1]
 xml2          1.2.2   2019-08-09 [1]
 source                                
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.3)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.3)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.3)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 local                                 
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        
 Github (bmschmidt/wordVectors@7f1914c)
 CRAN (R 3.6.0)                        
 CRAN (R 3.6.0)                        

[1] /Users/Primahadi/Rlibs
[2] /Library/Frameworks/R.framework/Versions/3.6/Resources/library

References

Biemann, Chris, Gerhard Heyer, Uwe Quasthoff & Matthias Richter. 2007. The Leipzig Corpora Collection: Monolingual corpora of standard size. In Matthew Davies, Paul Rayson, Susan Hunston & Pernilla Danielsson (eds.), Proceedings of the Corpus Linguistics Conference. University of Birmingham, UK. http://ucrel.lancs.ac.uk/publications/CL2007/paper/190_Paper.pdf (5 March, 2014).

Clark, Stephen. 2015. Vector space models of lexical meaning. In Shalom Lappin & Chris Fox (eds.), The Handbook of Contemporary semantic theory, 493–522. Second Edition. Hoboken: John Wiley & Sons.

Desagulier, Guillaume. 2017. Corpus Linguistics and Statistics with R. Cham: Springer International Publishing. doi:10.1007/978-3-319-64572-8.

Galili, Tal. 2015. Dendextend: An R package for visualizing, adjusting, and comparing trees of hierarchical clustering. Bioinformatics. doi:10.1093/bioinformatics/btv428.

Gries, Stefan Th. 2013. Statistics for linguistics with R: A practical introduction. 2nd ed. Berlin: Mouton de Gruyter.

Heylen, Kris, Thomas Wielfaert, Dirk Speelman & Dirk Geeraerts. 2015. Monitoring polysemy: Word space models as a tool for large-scale lexical semantic analysis. Lingua 157. (Polysemy: Current Perspectives and Approaches). 153–172. doi:10.1016/j.lingua.2014.12.001.

Hilpert, Martin & Florent Perek. 2015. Meaning change in a petri dish: Constructions, semantic vector spaces, and motion charts. Linguistics Vanguard 1(1). doi:10.1515/lingvan-2015-0013.

Larasati, Septina Dian, Vladislav Kuboň & Daniel Zeman. 2011. Indonesian Morphology Tool (MorphInd): Towards an Indonesian Corpus. In, Systems and Frameworks for Computational Morphology, 119–129. Springer, Berlin, Heidelberg. doi:10.1007/978-3-642-23138-4_8.

Levshina, Natalia. 2014. Geographic variation of Quite + ADJ in twenty national varieties of English: A pilot study. Yearbook of the German Cognitive Linguistics Association 2(1). 109–126. doi:10.1515/gcla-2014-0008.

Levshina, Natalia. 2015. How to do Linguistics with R: Data exploration and statistical analysis. John Benjamins Publishing Company.

Maechler, Martin, Peter Rousseeuw, Anja Struyf, Mia Hubert & Kurt Hornik. 2018. Cluster: Cluster Analysis Basics and Extensions.

Mikolov, Tomas, Kai Chen, Greg Corrado & Jeffrey Dean. 2013. Efficient Estimation of Word Representations in Vector Space. http://arxiv.org/abs/1301.3781 (14 December, 2018).

Mikolov, Tomas, Ilya Sutskever, Kai Chen, Greg Corrado & Jeffrey Dean. 2013. Distributed Representations of Words and Phrases and their Compositionality. http://arxiv.org/abs/1310.4546 (14 December, 2018).

Mikolov, Tomas, Wen-tau Yih & Geoffrey Zweig. 2013. Linguistic Regularities in Continuous Space Word Representations. In, Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 746–751. Atlanta, Georgia: Association for Computational Linguistics. http://www.aclweb.org/anthology/N13-1090 (14 December, 2018).

Nomoto, Hiroki, Hannah Choi, David Moeljadi & Francis Bond. 2018. MALINDO Morph: Morphological dictionary and analyser for Malay/Indonesian. In, Proceedings of the LREC 2018 Workshop "The 13th Workshop on Asian Language Resources", 36–43. http://lrec-conf.org/workshops/lrec2018/W29/pdf/8_W29.pdf.

Nomoto, Hiroki & David Moeljadi. 2019. Linguistic studies using large annotated corpora: Introduction. (Ed.) Hiroki Nomoto & David Moeljadi. NUSA 67. (Linguistic Studies Using Large Annotated Corpora). 1–6. http://repository.tufs.ac.jp/handle/10108/94450 (1 April, 2020).

Perek, Florent. 2016. Recent change in the productivity and schematicity of the Way-construction: A distributional semantic analysis. Corpus Linguistics and Linguistic Theory. doi:10.1515/cllt-2016-0014.

Quasthoff, Uwe & Dirk Goldhahn. 2013. Indonesian corpora. (Technical Report Series on Corpus Building). Leipzig, Germany: Abteilung Automatische Sprachverarbeitung, Institut für Informatik, Universität Leipzig. http://asvdoku.informatik.uni-leipzig.de/corpora/data/uploads/corpus-building-vol7-ind.pdf (26 July, 2015).

Rajeg, Gede Primahadi Wijaya, Karlina Denistia & Simon Musgrave. 2019a. R markdown notebook for vector space model and the usage patterns of indonesian denominal verbs. figshare. doi:10.6084/m9.figshare.9970205. https://figshare.com/articles/R_Markdown_Notebook_for_i_Vector_space_model_and_the_usage_patterns_of_Indonesian_denominal_verbs_i_/9970205.

Rajeg, Gede Primahadi Wijaya, Karlina Denistia & Simon Musgrave. 2019b. Vector space models and the usage patterns of indonesian denominal verbs: A case study of verbs with meN-, meN-/-kan, and meN-/-i affixes. (Ed.) Hiroki Nomoto & David Moeljadi. NUSA 67. (Linguistic Studies Using Large Annotated Corpora). 35–76. http://repository.tufs.ac.jp/handle/10108/94452 (1 April, 2020).

Rajeg, Gede Primahadi Wijaya, Karlina Denistia & Simon Musgrave. 2019c. Dataset for vector space model and the usage patterns of indonesian denominal verbs. figshare. doi:10.6084/m9.figshare.8187155. https://figshare.com/articles/Dataset_for_i_Vector_space_model_and_the_usage_patterns_of_Indonesian_denominal_verbs_i_/8187155.

Schmidt, Ben & Jian Li. 2017. wordVectors: Tools for creating and analyzing vector-space models of texts. http://github.com/bmschmidt/wordVectors.

Sneddon, James Neil, Alexander Adelaar, Dwi Noverini Djenar & Michael C. Ewing. 2010. Indonesian reference grammar. 2nd ed. Crows Nest, New South Wales, Australia: Allen & Unwin.

Wickham, Hadley & Garrett Grolemund. 2017. R for Data Science. Canada: O’Reilly. http://r4ds.had.co.nz/ (7 March, 2017).