=Paper=
{{Paper
|id=Vol-2870/paper16
|storemode=property
|title=Meiosis and litotes in The Catcher in the Rye by Jerome David Salinger: Text Mining
|pdfUrl=https://ceur-ws.org/Vol-2870/paper16.pdf
|volume=Vol-2870
|authors=Marta Karp,Nataliia Kunanets,Yuliia Kucher
|dblpUrl=https://dblp.org/rec/conf/colins/KarpKK21
}}
==Meiosis and litotes in The Catcher in the Rye by Jerome David Salinger: Text Mining==
https://ceur-ws.org/Vol-2870/paper16.pdf
Meiosis and litotes in The Catcher in the Rye by Jerome David
Salinger: text mining
Marta Karpa, Nataliia Kunanetsa,b and Yuliia Kuchera
a
Lviv Polytechnic National University 12 Bandera street, Lviv, 79013, Ukraine
b
Ivan Franko National University of Lviv, Universutetska Street 1, Lviv, 79000, Ukraine
Abstract
This paper deals with the different methods, particularly statistical analysis and text mining,
which help in stylistic research. The examination of the lexical and semantic features of
meiosis and litotes in the novel The Catcher in the Rye by Jerome David Salinger is presented
as an example. The examination in question has been carried out with the help of the
programming language R. To have a good-quality research, the specific features of litotes
and meiosis have been explored thoughtfully. Therefore, the broad range of possible
scientific views has been described, and, subsequently, we have made a general assumption
of typical linguistic patterns of meiosis and litotes. Using the obtained insights, it is possible
to apply different tools of text mining in stylistic research. The present paper outlines in
detail the creation of concordances, word frequencies and sentiment analysis. To reach our
goal, we have used the programming language R and the R packages which are distributed by
members of the community. In the scope of concordances, the concept of Key Word in
Context has been discussed as well, and the advantages of using concordances in stylistic
research have been introduced. The possible implementation of statistical analysis in the
research of litotes has been proposed and discussed. Within the framework of sentiment
analysis, we have focused on the negation, and how it affects the opinion orientation. Thus,
the present paper also aims to validate the importance of litotes in sentiment analysis, as
litotes are directly linked to the effects of negation. The results of each stage of the research
have been provided and meticulously discussed.
Keywords1
Meiosis, litotes, natural language processing, text mining, sentiment analysis
1. Introduction
The present research consists of several parts, namely the statistical analysis, the basic methods
and instruments of text mining and the semantic-stylistic analysis. Therefore, the work includes both
technical and philological ways of working with texts, particularly with the novel The Catcher in the
Rye by Jerome David Salinger. A significant part of the present paper is dedicated to the more
linguistic aspects of the data under the study, which are lexical and semantic features of meiosis and
litotes. More than that, we have explored in which way the statistical analysis is helpful during these
kinds of research, and how the litotes impact the machine processing of the texts.
2. Theoretical framework
The present paper mainly deals with the ways in which we have applied statistical approaches and
methods of text mining for stylistic investigations. We are focusing on such stylistic devices as
COLINS-2021: 5th International Conference on Computational Linguistics and Intelligent Systems, April 22–23, 2021, Kharkiv, Ukraine
EMAIL: martakarp26@gmail.com (M. Karp); nek.lviv@gmail.com (N. Kunanets); yuliia.s.kucher@gmail.com (Yu. Kucher)
ORCID: 0000-0002-7332-7739 (M. Karp); 0000-0003-3007-2462 (N. Kunanets); 0000-0002-0728-6617 (Yu. Kucher)
©️ 2021 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
meiosis and litotes in particular. That is why first and foremost it is important to discuss the
theoretical aspects of those devices. Both meiosis and litotes are the types of understatement,
however, there are several significant distinctions between them. Therefore, we suggest a brief
overview to better understand their specific features. So, meiosis stands for the figure of speech,
which refers to the object in a special way, therefore deliberately reducing its significance.
Commonly, they are functioning in the sentence as modifiers, and, subsequently, they are such parts
of speech, as adjectives or attributive nouns. On other hand, litotes expresses the meaning of the
understatement by denying some qualities to claim the opposite. So, litotes takes the form of “the
negation of something”, and on the syntactic level it generally embraces the next pattern:
“not/no/none + something”; however, it is not limited to one. To clarify all the specific features of
meiosis and litotes, we suggest to look further into the ideas of different scholars, who profoundly
have worked with these stylistic devices. Hence, James Jasinski proposes such a definition of meiosis:
“Meiosis is a statement that depicts something important in terms that lessen or belittle it”. Besides,
he states that: “Any verbal effort to make an event, an idea, or a person less significant is a form of
meiosis” [9, p. 550]. Another researcher, Brett Zimmerman, who is interested in Edgar Allan
Poe’s style, offers the following description of the meiosis: “a lessening, sometimes belittling thing or
person, possibly with a degrading epithet, or with the substitution of a word.” [26, p. 43]. Very often
the stylistic meaning of meiosis is reduced only to the ironic effect of the understatement, and we
observe it in the definition of Bernard Marie Dupriez: “A figure which uses ironic understatement to
represent something as in some way less than it is: a form of ironic emphasis.” [3, p. 273]. A very
comprehensive definition is given by Aida Besancon Spencer in her stylistic study of the Apostle
Paul’s style of written communication. She writes the following: “Quintillian says that meiosis may
refer to a style as well as a figure. 1. When applied to a style, meiosis indicates “meagreness and
inadequacy of expression,” characterizing an ‘obscure style rather than one which lacks ornament.’”
2. Meiosis and litotes are often not distinguished. Meiosis is a deliberately employed understatement,
presenting something as less than it really is. It is belittle, often through a change of meaning of one
word, as in using a degrading epithet [20, p. 195].
Among the most prominent scholars, who worked with litotes, we draw special attention to Otto
Jespersen, Laurence R. Horn, Dwight Bolinger and Ton van der Wouden. We have picked exactly
their works, as we believe that their views and ideas are perfectly suited to our research. Jespersen in
his work Negation in English and Other Languages, describes the cases of double negation and states
the following: “It seems to be a universal rule in all languages that two negatives make an affirmative,
if both are special negatives attached to the same word; [...] But it should be noted that the double
negative always modifies the idea, for the result of the whole expression is somewhat different from
the simple idea expressed positively.” [10, p. 63]. Ton van der Wouden provides us with a rather
comprehensive analysis of the nature of litotes, where he, inter alia, refers to the Horn’s analysis of
litotes. Therefore, Wouden describes, that Horn’s analysis of litotes has two parts: a semantic one and
a pragmatic one. The semantic one embraces the ideas that two negations make an affirmative. The
pragmatic one has been explained in the following interpretation of one of Grice’s maxims, and is
called Division of Pragmatic Labour, meaning: “The use of a longer, marked expression in lieu of a
shorter expression involving less effort on the part of the speaker tends to signal that the speaker was
not in a position to employ the simpler version felicitously.” [24, p. 122-123]. Horn, in his paper
Duplex negatio affirmat...: The Economy of Double Negation writes that “The expectation that two
negatives SHOULD cancel out is a linguistic reflex of the logical Law of Double Negation (LDN),
~(~α) ≡ α” [8, p. 80]. However, as he elaborates further on this matter, he suggests that there are many
nuances. Thus, he cites Bolinger and his thoughts on litotic negation: “... the denial of the negative
leaves the entire positive range open to whatever degree is appropriate. The litotes, in fact, call
attention to this gradient – the hearer is invited to consider the degree to which the facts point.” [2,
p. 116]. The latter sentence is indeed overlapping with the Division of Pragmatic Labour, implying
that litotes is stylistically less natural and more complex, and that the speaker usually uses the double
negative (or the negative of the contrary) in lieu of simple positive description is quite deliberate. So,
moving forward on Wouden’s description, we have seen that he describes the so-called “grey zone”
between one world and its negation. For example, in the next sentence: It’s not too bad when the
sun’s out, but the sun only comes out when it feels like coming out [17, p. 202].
Here we see such litotes as not too bad and it is quite common in the English language. Taking
Wouden’s discourse as a basis, we have derived the following: the opposition to “bad” is “good”,
however, these words are not contradictory, but contrary or gradable antonyms, meaning that they do
not refer to absolute qualities, therefore, they are not strict oppositions. The contrary antonyms are
pictured as a scale with these two words as polar members, and what is the most important, they admit
possibilities between them. In our case, the more direct word “good” is deliberately omitted in lieu of
negating the opposite word “bad”. Thus, the specific range between two meanings of “good” and
“bad” is created, where “not bad” tends to be closer to “good”, but, and it is the main point there, does
not exactly substitutes it. It is pictured in the following scheme:
somewhat
too bad somewhat bad too good
good
not too bad
So, we have observed an area between two extremes too bad and too good, where neither of them
applies. As Bolinger writes: “When intensifiers are present, the litotes tends to deny one end of a
polarity to imply an encroachment on the other end. [...] The negative passes conceptually from the
intensifier to the intensified, with the intensifier weakened to rather” [2, p. 116]. Thus, not too bad
corresponds with rather good. However, Bolinger also mentions, that there is a discrepancy between
the litotes with an intensifier, and without one. He calls the litotes without an intensifier a
contradictory one – it expresses that the entire opposite range is open. For instance, if we are to
examine the litotes not unwilling, the results are presented in the following scheme:
willing ... unwilling
all that is not unwilling
where not unwilling embraces any degree of willingness. On the other hand, Wouden presents a
slightly different approach to the analysis of litotes. Drawing upon his notions, we have inspected
litotes not unwilling in the following way:
1. willing ... unwilling
2. not willing
3. not unwilling
4. not unwilling
Thus, according to the Wouden, in the first row, we have seen two extremes, willing and unwilling,
and the space between them, and it is exactly the grey zone, that has been mentioned before. It is the
zone, where neither willing, nor unwilling is not applicable. We also have to keep in mind the fact,
that there is no definite boundary between the boundary and the grey zone. The second row pictures
what Wouden calls logical denotation of not willing. Logically, the negation of willing covers the
space on the scale that remains. Similarly, with the third row, logical denotation of not unwilling is the
rest of the scale, but unwilling. However, as we have seen in the fourth row, the pragmatic denotation
of not unwilling opens the only possible range that is between the two extremes. Thus, we have made
the next interim conclusion: not unwilling refers only to the middle part of the scale, so it is not direct
opposition to the unwilling, but a litotes which carries the effect of understatement. Moving forward
to the next important details about the features of litotes, both Horn and Wouden claim the following:
“Litotetic constructions with nongradable predicates are ungrammatical or unfelicitous, equivalent to
the straightforward expression, or figurative by necessity.” [24, p. 123]. Horn, in turn, proposes, that
“[...] litotes, so defined, does not require single, let alone, double negation.” [8]. He also proposes that
not all examples of double negations are obligatory litotes. We pay attention to this idea, as it has
become clear that we are not entitled to assume that every pattern of double negation has the stylistic
effect of litotes. So, Horn’s point of view can be pictured in the following diagram:
Another important matter with the negation of the contrary is that it is utterly context, and
sometimes even intonationally, dependent. We have already mentioned, that not bad is a quite popular
example of litotes. However, if we take a look at the following extract, we see that the function of the
phrase is not litotic:
He didn’t know what the hell I was talking about, so all he said was “Oh” and took me up. Not
bad, boy. It’s funny. All you have to do is say something nobody understands and they’ll do
practically anything you want them to [17, p. 205].
As we have determined from the context, not bad does not act as underestimation. Its functions are
exactly opposite – it has the meaning of overestimation, which can be rendered as very good. Stern in
Horn [7, p. 356] writes the next statement about litotes: “Not bad, taken literally,
leaves a large latitude, from indifferent to excellent, and may mean [sic]
either, depending on the intonation used and the circumstances”. Bolinger [2, p. 116], in turn, states
that “A familiar example is the interjection Not bad, usually written with an exclamation point to
indicate the intonation of surprise that suggests Very good!, but without to indicate the terminal fall-
rise that damns with faint praise.”
Statistical data and quantitative analysis are of great importance in nowadays linguistical research.
They help the scientists to get clear and objective results, and it is a high priority in the realm of
science. Thus, we are also working with the methods that are parts of the general scope of statistics
for linguistics. We want our methods to be agile, veritable, and precise, so we believe that the best
tool that satisfies all requirements, is programming language R. R is one of the most popular tools
among statisticians and data miners, on the one level with such well-known languages as Python and
MATLAB.
The workflow of analysing data consists of several steps:
Accessing the data (we store information for analysis in different places, so inevitably we
need to get the data into our application)
Cleaning the data (the data may be kept in such formats, that is not appropriate for the
analysis; also, some parts may be missing or/and miscoded, so we should take care of it)
Annotating the data (make the relations between the pieces of data and what they represent)
Summarising the data (statistical characterisation of the data)
Visualization the data (the possibility to create different graphs, plots, histograms, charts etc.)
Modelling the data (applying the mathematical and statistical rules to the set of data to
identify relationships between its parts and accordingly make hypotheses for further
investigations)
Preparing the results (presenting the outcomes in the publication-quality and user-friendly
formats) [11, p. 24]
The list above is the most general way to think about the data analysis. The R’s capability allows
us to accomplish all of these steps, so this is our starting point from the statistical point of view.
Another definition, which has been used often in the present paper is Text Mining. This term embraces
the large scale of different concepts and practices; however, it is important to describe it generally to
ensure the clear comprehending of the further explanations, and to avoid ambiguity. Therefore, as Ian
H. Witten states in the paper concerning text mining [25], this term is described as attempts to glean
meaningful information from natural language text. The term text mining is confused with data
mining, although they are quite different notions. The latter one is all about searching for patterns in
data, and therefore its main goal is to retrieve the comprehensible information from the large amount
of input data. What is also important to mention, is that the input data in data mining is implicit,
meaning that raw data make no or little sense for human comprehension, and is only analysed with the
help of different automatic techniques of data mining. The main problem of text mining is different.
In the text mining, the input data is almost perfectly clear for human, as it is texts written in natural
language. However, for the machine processing, the input data which has the form of the texts, might
be even more complicated, than data from different databases. Working with text mining, you are
indispensably to encounter the notion of Natural Language Processing, or NLP. To eliminate the term
above, we are citing the definition, introduced by Elizabeth D. Liddy: “Natural Language Processing
is a theoretically motivated range of computational techniques for analysing and representing
naturally occurring texts at one or more levels of linguistic analysis for the purpose of achieving
human-like language processing for a range of tasks or applications.” [12]. At the present time, there
are many various applications of NLP, for example, machine translations, automatic summarization,
sentiment analysis, conversational agents (chatbots), and so on. In the present paper, we decide to use
the following methods and techniques: word frequency, collocation, concordance and sentiment
analysis.
One of the ways, in which we have explored the possibilities for text mining (or text analysis), is
building the concordances. The concordance is a set of the searched words (or collocations), within
the context, and they are utterly useful for the investigation of meiosis and litotes, especially in
fiction. The concordance is best perceived in the form of KWIC (Key Word in Context), but the ways
of presenting the outlines are not limited to KWIC. As Stefanowitsch [21, pp. 50-51] has rightly
stated, KWIC concordance means displaying the hits for the query in the proximate context, so there
are a particular number of words or characters to the left and the right of it. He also mentions that
concordance provides an overview of the typical usage of word forms (or sets) that is of our interest.
Hence, concordances in the form of KWIC have been presented in the paper.
A vitally important step in text analysis is preprocessing. This process is all about data cleaning,
and it is crucial to complete this step in a correct way because with “dirty” data it is very hard to
undertake proper analysis. In the realm of text data cleaning means several actions:
Removing the punctuation, such as period, exclamations points, question marks, commas,
semicolons, colons, hyphens etc.
Changing all the words to the lower case
Removing the numbers
Removing the stop words
Concerning the last point in the list above, it might be necessary to elaborate on the concept of stop
word. Traditionally, in the realm of text mining, the stop words are the words that bear no semantic
value, and therefore have no or little impact on the overall text perception. Stop words are usually
articles, prepositions, conjunctions and pronouns. For example, such words as a, the and it provide
rather scarce information and low semantic significance. These stop words just confuse the researcher
and may cause unnecessary problems, meaning that they are redundant, so it is advisable to get rid of
them [6]. However, it is crucial to keep in mind, that the removal of stop words not always is
considerate to be the best practice. There are cases where the removal of stop words can corrupt the
process and therefore the results of the investigation. The case, which directly concerns the present
research, is outlined as follows: the removal of stop words causes changes in the meaning of the
undertaken text. For example, the removal of words with the meaning of negations, such as not, no,
non, transforms the meaning of the sentence into exactly the opposite one. After the cleaning, in the
sentence “It is not a good thing”, the only meaningful words for computer programme are good and
thing, and it is analysed as something with positive connotations, however, we understand, that the
original sentence has rather negative connotation.
Another important matter of the present paper is sentiment analysis. As Bing Liu [13] states in his
paper Sentiment Analysis and Subjectivity: “Sentiment analysis or opinion mining is the
computational study of opinions, sentiments and emotions expressed in text”. So, this specific type of
analysis is all about the processing of people’s opinions. The development of the Internet has resulted
in the establishing of a new unique space, where people have become able to express their opinions in
various ways (e.g., reviews, comments, blogs). For the last years, sentiment analysis has come a long
way of development, and we distinguish the set of the most common features, that used in research
and practice:
Terms and their frequency (generally well-known technique in text analysis that is used in
sentiment classification as well)
Part of speech tags (the prevailing part of speech that is responsible for indicating the
subjective opinions is an adjective)
Opinion words and phrases (opinion words are the words with positive or negative
connotations; for instance, the word beautiful has positive evaluation, while the word rubbish
expresses clearly negative sentiment)
Syntactic dependency (the word positions in sentences also influences the way that sentence
is perceived)
Negation (negations often change the opinion orientation, however, there are cases when the
negative particles do not mean negation, as, for example, in pattern not only ... but also)
The main concern in the present study is a litotes and meiosis, thus, we have inspected the
sentiment analysis through the prism of these stylistic devices. What is interesting here, is the
methods, with the help of which, sentiment analysis deals with double negation and litotes. Professor
Liu discusses the effects of negation in opinion mining, and outlines the next effects of negation:
Negation Neg → Positive
Negation Pos → Negative
These insights are quite logical, however, as we have mentioned before, the litotes is the
understatement that is resulted from the double negation or the negation of the contrary. The negative
particle in litotic construction does not state the direct opposite, but a weaker meaning. We will
discuss the litotic effect in the sentiment analysis further in Section 3.3.
3. Practical analysis
The present study is dedicated to two major problems: how statistical data and data mining help
during stylistic research, and how stylistic devices, especially litotes, impact the sentimental analysis.
The main tool of our investigation has been R, which is a language and environment for statistical
computing and graphics. We have decided to use this particular technology infrastructure, as it allows
us to implicate various data analytic techniques. R provides a range of different instruments, starting
from quite easy ones, that do not require special knowledge, and ending with advanced powerful
tools. More than that, R have a very useful feature called packages. To elucidate this term, it is
necessary to refer to Robert Kabacoff [11, p. 54]: “Packages are collections of R functions, data, and
compiled code in a well-defined format.” In other words, packages are shareable bundles of code.
Thus, there are a huge community of other R users, who have already dealt with the same problems
that may occur during our work, so with the help of packages it is possible to benefit from their
contributions, and significantly optimise your workflow. This feature is what makes R such a popular
tool for undertaking different investigation. More than saving your time, it also enhances the process
of the research in a way, that makes the investigation more agile. What we mean by that, is that the
investigation is divided into several steps, and it enables us to manage every step, hence, you gain
more control over what you do. So, we believe that R is a perfect and balanced tool for such linguistic
investigations, like ours.
3.1. Concordance
The building of the concordance is an important part of linguistic analysis. It enables rapidly and
efficiently process a vast amount of data. Usually, the results are displayed in the form of keyword-in-
context displays or KWICs. This form of KWIC implies the searched word (or phrase) is shown in
context, meaning that there is a couple of words before and after the searched word or phrase. For our
analysis, we will use the package quanteda [1]. This package is used for managing and analysing text,
and it also contains a very useful function kwic. It is quite easy to use, nevertheless, it is agile and
provide with sufficient number of ways, in which the output concordance is presented. As our study is
about meiosis and litotes, as well as their semantical and lexical features, the concordances give us
insights that are quite important. Meiosis is a stylistic device with the functions of lessening the
significance of something, thus, we have made an assumption that such words as little, small and tiny
could be used. Therefore, to build a concordance, firstly we need to access the data. It has been done
with the help of tm package [4]. Overall, tm package is a great framework for text mining
applications, which also contains several useful functions, that we are going to apply later. The main
structure for managing documents in tm is a Corpus, representing a collection of text documents. Tm
also contains a function with the same name. As expected, the Corpus function creates a corpus. The
first argument to Corpus is in which way we want to create the corpus. In our case, we have read PDF
file from the directory with the help of the built-in function DirSource. Using the second
argument, readerControl, we tell Corpus which reader to use to read the text from the PDF files. That
has been readPDF, also a tm built-in function. The readerControl argument requires a list of control
parameters, one of which is reader. After these manipulations, what we have got is a corpus, that in
our case consists only of one document. Subsequently, we have used function kwic, that allows us to
easily extract concordances. The kwic function takes the text (x) and the search pattern (pattern) as its
main arguments but it also has some additional arguments, that helps to make the concordance more
flexible. The simplest way of creating concordance with the searched word tiny is pictured in
Figure 1.
Figure 1: Concordance with word tiny
With the help of the concordance, we have distinguished such example of meiosis, as, for instance,
tiny little tumor. The whole extract is the following:
“It isn’t very serious. I have this tiny little tumor on the brain.”
“Oh, no!” She put her hand up to her mouth and all. “Oh, I’ll be all right and everything! It’s
right near the outside. And it’s a very tiny one. They can take it out in about two minutes.” [17, p. 75]
The next sentences after the tiny little tumor even enhance the effect of the meiosis. Thus, in such a
way, we can investigate other occurrences of meiosis as well.
As for litotes, we have created concordances for them too. The most common patterns for litotes
are double negation and negation of the contrary. We will use the same kwic function, however, at this
step we have added extra arguments: window and valuetype. The former one specifies how many
words/elements are shown to the left and right of the keyword. By default, there are 5 words by each
side of the searched term, and for a better understanding of the context, we have expanded it to 10
words. The valuetype argument is responsible for the type of pattern matching: “glob” for “glob”-
style wildcard expressions; “regex” for regular expressions; or “fixed” for exact matching. For
example, we can create a concordance for the pattern “not + word with negative prefix”, as in
Figure 2.
Figure 2: Concordance with litotic pattern
We can investigate litotes not bad in the context accordingly. It is also important to mention that
there can be other words, intensifiers in particular, between not and bad, thus, we can use regular
expressions to achieve our goal (Figure 3).
Figure 3: Concordance with litotes not bad
As we see, there are examples of litotes with intensifiers between the negative particle and the
main word. Such cases are important to the same extent as the classical examples of litotes.
To sum up, concordances can provide a very convenient way of investigating the cases of litotes
and meiosis in the context. From the philological point of view, it allows us to see and investigate the
examples of the rhetorical devices. The concordances demonstrate the convenient way of analysing
meiosis and litotes in the scope of the novel, as we have operated not only with the numbers of found
examples but also immediately have seen the context and have regulated it by choosing the quantity
of words or symbols before and after the searched keyword. This approach is quite convenient
because it is almost impossible to undertake the appropriate analysis of the stylistic devices without
the context. Regarding the application of this method to the novel The Catcher in the Rye, some
results have been outlined as follows: we have observed some examples of the meiosis and litotes in
the text, however they are not numerous. Nevertheless, they still significantly contribute to the
stylistic perception of the novel. It is also worth mentioning, that due to the specific recognisable
syntactic pattern of litotes, they are more easily detectable via the text mining tools. More than that,
these obtained concordances will serve as a basis for further research.
3.2. Word frequencies
Just like with the task of creating concordance, first of all, we have extracted a text in a format that
is suitable for R. Therefore, we need a package that allows us to work with PDF format in R. Such
package exists and it is called pdftools [15]. The pdftools function for extracting text is pdf_text. As
result, we have a vector that contains the text of the PDF file. The length of a vector corresponds to
the number of pages in the PDF file. In our original PDF file, we have 116 pages, hence, the length of
our vector is 116.
With the help of such packages as tidytext [18], tibble [14], dplyr [23] and ggplot2 [22], it is
possible to undertake an assessment of words and n-grams frequencies, and, finally, to build plots
with the results. Thus, recalling the steps of data analysis flow, we have transformed our raw data into
tidy text format. Getting your text into tidy text format is believed to be a fundamental requirement to
perform text mining and counting word frequencies. Therefore, we have changed the already
transformed text of PDF format into tibble (data frame). A data frame is a two-dimensional array,
where different columns contain different modes of data (numeric, character, and so on). There are
differences between tibbles and data frames, but they are insignificant in the current study. After that,
we have used the unnest_token function from tidytext that, roughly speaking, performs the
tokenization. Tokenization, in NLP, is a process of transforming the human-readable text into
machine-readable data. There are various ways of tokenization, the most common one is splitting the
text into separate words. In the present study, we have used this method, as well as breaking the text
into n-grams. N-grams are described as a consecutive sequence of words, where a bi-gram is a pair of
two words, a tri-gram is a group of three words and so on. Moreover, the unnest_token function
removes all punctuation and converts words to lower case. One more significant step is removing stop
words. It is a common recommendation for creating tidy text formats, however, the negative words
like not, no, none are considered stop words, but they are vital for our analysis, therefore we have
skipped this step. Finally, using a function filter, we have counted and compared the frequency of
occurrence of word combination not bad, which includes the features of litotes and word good with a
direct meaning. Similarly, like in the case with concordances, it is crucial to remember about
intensifiers, that occur between not and bad. We have visualised the results with the help of the
ggplot2 package and its functions (Figure 4).
Figure 4: Frequencies of the word combination not bad and good
The plot has shown us the significant difference between frequencies of word combination not
bad, word combination not bad with the intensifier, and good. Despite the existence of several litotic
cases, it is clear that the non-litotic is prevalent for this case.
The same comparison can be made for the opposite – word combination not good, word
combination not good with the intensifier, and bad.
Figure 5: Frequencies of the word combination not good and bad
Different stylistic devices are an important part of the story. They enrich the language and attach a
unique style to the text. As for litotes, they are especially important, as they lessen the significance of
the statement in a special way, therefore provoke the reader to contemplate the situation. However,
after analysing the concordances and word frequencies for this stylistic device, we conclude that the
author has decided to use more direct forms instead of litotic constructions. Such peculiarity of usage
of this type of understatement is one of the many others features of Salinger’s idiostyle.
3.3. Sentiment analysis
There are several approaches to sentiment analysis. The first one is based on unigrams (or single
words) and the second one is based on sentences. In other words, the difference is in units of
tokenization – the former is done at the word level, while the latter is done at the syntactical level. The
sentiment analysis that is undertaken with the help of single word tokenization is fruitful, however, there
are cases when it can distort the facts. For instance, if we take the “afinn” lexicon from the tidytext
package, which assigns words with a score that is in a range between – 5 and 5, with negative scores
indicating negative sentiment and positive scores indicating positive sentiment. We have tried to
undertake the analysis of sentence with litotic construction, using the “afinn” lexicon (Figure 5): I mean
it isn’t too nice, naturally, if somebody tells you you don’t brush your teeth [17, p. 32].
Figure 6: Results of “afinn” sentiment analysis
After the analysis, we have got that the sentence has only one word which corresponds with
“afinn” lexicon, and its evaluation is 3 points. It does not properly reflect the sentiment of the
sentence. For such incidents, we have used the tools which undertake the tokenization on the level
beyond the single word. One of such tools is sentiment [16]. It has been used for the calculation of the
text polarity sentiment at the sentence level. Thus, after we get the text in an appropriate format, we
have used the built-in functions get_sentences, sentiment_by and highlight. The first one is used for
breaking the text into sentences. The second one actually performs the sentiment analysis. Finally, the
last one allows to highlight positive/negative sentences as an HTML document (positive = green;
negative = pink). The result is presented in Figure 6 (only one part of a whole text).
Figure 7: Results of the sentiment analysis
The sentence that we have discussed before has a negative evaluation. Thus, we have inferred that
the analysis is correct. However, the litotes in the following sentence: “He’s not too bad,” I said.
“You don’t know him, that’s the trouble.” [17, p. 32] is assessed as negative, while from the
theoretical background, section 2, we have outlined that not too bad is corresponds with rather good.
These outcomes show that some tools for sentiment analysis are better if we want to take into
account the stylistic effect of the litotes. However, the litotic impact is considerably context-
dependent, so, even if we are trying to analyse them in the scope of not a single word, but a whole
sentence, the results still are not ideal.
4. Conclusions
In this paper, we have thoroughly examined some ways of implementing text mining and statistical
methods into stylistic research. As a basis, we have chosen such stylistic devices as meiosis and litotes
in The Catcher in the Rye by Jerome David Salinger.
The whole research has consisted of several constituent stages. Firstly, we have highlighted the
important notions concerning the lexical and semantical characteristics of litotes and meiosis. We
have also outlined the most general syntactical patterns of litotic expressions and examined the
concepts of the negation of the opposite and double negation at this stage. The main principles of
analysing data have been introduced as well. In the scope of preprocessing, we have examined the
stop words and their impact on the quality of the results. We have also drawn our attention to the
notions of sentiment analysis, and the role negation plays during the performance of such opinion
assessment analysis. At the second stage of the research, we have undertaken the practical analysis of
namely the creation of the concordances, the word frequencies and sentiment analysis. The
programming language R and the R packages have been used as a tool for examination. The
significant part of this stage has been the correct and demonstrable visualisation of the outcomes.
The results have shown us that we can successfully use the R tools to get valuable stylistic
insights. The concordances are a great way of investigating the typical examples of meiosis and
litotes in context. The word frequencies of litotic structures and their non-litotic correspondences, and
their correlations have been visualized and thus provide us with the overall understanding of the
functioning of litotic expressions in the text. Finally, we have investigated different sentiment
analysis’ approaches, and conclude how each of them treats the litotes.
The present paper potentially leads to other researches in this field. Our work suggests that more
investigations can be undertaken in the scope of text mining and stylistics. One of the possible
directions can be related to the problem of the significance of litotes and meiosis in the text of
different genres. For example, the correlation of occurrences of these stylistic devices throughout the
range of various texts can be investigated.
5. References
[1] K. Benoit, K. Watanabe, H. Wang, P. Nulty, A. Obeng, S. Müller, A. Matsuo, quanteda: An R
package for the quantitative analysis of textual data, Journal of Open Source Software 3(30)
(2018). URL: https://quanteda.io. doi: 10.21105/joss.00774.
[2] D. Bolinger, Degree Words, Mouton, The Netherlands, 1972.
[3] B.M. Dupriez, A Dictionary of Literary Devices: Gradus, A-Z, University of Toronto Press,
Toronto and Buffalo, 1991.
[4] I. Feinerer, K. Hornik, tm: Text Mining Package. R package version 0.7-8, 2020. URL:
https://CRAN.R-project.org/package=tm.
[5] I. Feinerer, K. Hornik, D. Meyer, Text Mining Infrastructure in R, Journal of Statistical
Software, 25(5) (2008) 1–54. doi: 10.18637/jss.v025.i05.
[6] E. Haddia, Xiaohui Liua, Yong Shib, The Role of Text Pre-processing in Sentiment Analysis,
Procedia Computer Science 17 ( 2013 ) 26 – 32.
[7] L. R. Horn, A natural history of negation, CSLI Publications, Stanford, 2001.
[8] L. R. Horn, Duplex negatio affirmat...: The Economy of Double Negation in L. M. Dobrin, L. N.
& R. M. Rodriguez (Eds.), CLS 27: Papers from the 27th Regional Meeting of the Chicago
Linguistic Society. Part Two: The Parasession on Negation, Chicago Linguistic Society,
Chicago, 1991, pp. 80-106.
[9] J. Jasinski, Sourcebook on Rhetoric, Sage Publications, Inc., California, 2001.
[10] O. Jespersen, Negation in English and Other Languages, Forgotten Books, London, 2012.
[11] R. Kabacoff, R in Action: Data analysis and graphics with R, 2nd. ed., Manning Publications
Co., 2015.
[12] E.D. Liddy, Natural Language Processing in Encyclopedia of Library and Information Science,
2nd. ed., Marcel Decker, Inc., NY, 2001.
[13] B. Liu, Sentiment Analysis and Subjectivity in: N. Indurkhya, F. J. Damerau (Eds.), Handbook of
Natural Language Processing, 2nd. ed.,Chapman & Hall, Boca Raton, FL, 2010.
[14] K. Muller, H. Wickham, tibble: Simple Data Frames. R package version 3.0.5., 2021. URL:
https://CRAN.R-project.org/package=tibble.
[15] J. Ooms, pdftools: Text Extraction, Rendering and Converting of PDF Documents. R package
version 2.3.1., 2020. URL: https://CRAN.R-project.org/package=pdftools.
[16] T. W. Rinker, sentimentr: Calculate Text Polarity Sentiment. R package version 2.7.1., 2019.
URL: http://github.com/trinker/sentimentr.
[17] J. D. Salinger, The Catcher in the Rye, Little, Brown and Company, Boston, 1951.
[18] J. Silge, D. Robinson, tidytext: Text Mining and Analysis Using Tidy Data Principles in R,
JOSS, 1(3) (2016). doi: 10.21105/joss.00037.
[19] J.M. Sinclair, Corpus, Concordance, Collocation, Oxford University Press, Oxford, 1991.
[20] A.S. Spencer, Paul's Literary Style: A Stylistic and Historical Comparison of II Corinthians
11:16-12:13, Romans 8:9-39, and Philippians 3:2-4:13, University Press of America, Inc.,
Lanham/Oxford, 1998.
[21] A. Stefanowitsch, Corpus Linguistics. A Guide to the Methodology. Textbooks in Language
Sciences, Language Science Press, Berlin, 2020.
[22] H. Wickham, ggplot2: Elegant Graphics for Data Analysis, Springer-Verlag, New York, 2016.
[23] H. Wickham, R. Francois, L. Henry, K Muller, dplyr: A Grammar of Data Manipulation. R
package version 1.0.3., 2021. URL: https://CRAN.R-project.org/package=dplyr.
[24] T. van der Wouden, Negative Contexts (Linguistics), Ph.D. thesis, University of Groningen,
Groningen, The Netherlands, 1994.
[25] I. H. Witten, Text Mining, University of Waikato, New Zealand, 2004.
[26] B. Zimmerman, Edgar Allan Poe: Rhetoric and Style, McGill-Queen’s University Press,
Montreal & Kingston, 2005.