Comparing Close and Distant reading outcomes of Virginia Woolf’s Mrs. Dalloway.

For the text analysis praxis, I chose to test out an observation made while close reading Virginia Woolf’s Mrs. Dalloway. For another DH class, I’ve been working on a group project investigating annotation methods using the Mrs. Dalloway as the basis. Despite Woolf suggesting that she simply let the novel flow through her with no great structure in mind, when reading it’s hard to not feel the seemingly concrete structure on top of which flows the seemingly meandering stream of consciousness dance. Of course that stream of consciousness surfaces themes again and again via the minds of various characters until it becomes plain that Woolf wants us to explore certain aspects of our humanity—youth and age, spontaneity and duty among them. 

When close reading I felt and remarked that the color grey turned out repeatedly. At first I suspected that the London setting and widely accepted stereotypical assumptions of English weather made me more attuned it (ahh yes, confirmation of that stereotype, I’m picturing the right thing) —maybe even made me overestimate its prevelance. But then I began to notice it being applied to all manner of things—but most often in a manner to denote stature, wisdom, age, and respectability. I also began to nice the mention of roses—a kind of sprinkling evenly throughout the text in relationship to various characters. These flowers were derided, gifted, and displayed—even used by Septimus, a character with a tenuous connection to this Earth, to ground himself in the moment and counteract his tendency lose himself to incoherent thought. I decided that Voyant would be a great way to see what the algorithm’s had to say about my observations. Would the reinforced or minimized—would it help suss out a pattern I didn’t observe?

Looking at the frequency of grey (38 mentions), which appears more than any other color in the text after white (58 mentions), I did see a continued usage throughout the text, and the context revealed what I had suspected.  It is often used to denote age (grey hair)—and suggest a more regular and fixed time of life. It is, also, used to signal a more refined air—standing in the world. In relation to the text, I’d even go so far as to say it showcased the kind of fixed and respectable striving of 1923 London. Weather does come in, for sure (grey-blue sky), but more often it describes the appearance of understated and refined clothing, vehicles, and homes of established ladies and gentlemen.

Indeed it was—Sir William Bradshaw’s motor car; low, powerful, grey with plain initials’ interlocked on the panel, as if the pomps of heraldry were incongruous, this man beingthe ghostly helper, the priest of science; and, as the motor car was grey, so to match its sober suavity, grey furs, silver grey rugs were heaped in it, to keep her ladyship warm while she waited.

He had worked very hard; he had won his position by sheer ability (being the son of a shopkeeper); loved his profession; made a fine figurehead at ceremonies and spoke well—all of which had by the time he was knighted given him a heavy look, a weary look (the stream of patients being so incessant, the responsibilities and privileges of his profession so onerous), which weariness, together with his grey hairs, increased the extraordinary distinction of his presence and gave him the reputation (of the utmost importance in dealing with nerve cases) not merely of lightning skill, and almost infallible accuracy in diagnosis but of sympathy; tact; understanding of the human soul. 

In considering roses, I wondered if it might not be more congruous, given my investigation of the color grey, to shift to investigating the color red. The roses mentioned throughout the book are red—that classic color of love, emotion, youth and intensity. When I made this adjustment something very striking was revealed. The frequency of the mention of red mirrored that of grey—almost as though they went hand in hand. Instead of black and white, Woolf seemed to have leaned into a contrast of grey and red—old vs young, passion vs resignation. 

The striking overlap is even more apparent when contrasted with the color most mentioned in the book— white. There is a clear correlation between grey and red.

Investigating red’s term berry doesn’t give too much away, but digging into the context gives everything away. It’s often mentioned in descriptions with almost riotous abundance of color (in stark contrast to drab and monotonous grey) and often paired with descriptions of flowers and bodily features (again with hair and clothing, but also nostrils, lips, and cheeks). Red is seemingly used to the character experience in the present moment of the novel in contrast to the experience of reminiscence that makes up much of the novel.  

… and it was the moment between six and seven when every flower—roses, carnations,
irises, lilac—glows; white, violet, red, deep orange; every flower seems toburn by
itself, softly, purely in the misty beds; and how she loved the grey-white mothsspinning
in and out, over the cherry pie, over the evening primroses!

But she’s not married; she’s young; quite young, thought Peter, the red carnation he had seen her wear as she came across Trafalgar Square burning again in his eyes and making her lips red.

…through all the commodities of the world, perishable and permanent, hams, drugs, flowers, stationery, variously smelling, now sweet, now sour she lurched; saw herself thus lurching with her hat askew, very red in the face, full length in a looking-glass; and at last came out into the street.

And that very handsome, very self-possessed young woman was Elizabeth,over there, by the curtains, in red. 

Over all this investigation made me curious about mapping and investigating the many many overlapping patterns and structures in the novel. In fact, this exercise and that in my other class has pushed me into a peculiar space of looking at the novel as a specimen to be poked, prodded, and labeled, graphed and displayed. In this way I might possess and express my own experience of it. The idea of thinking one could “master” a novel in this way feels a bit like a delusion. It’s like trying to create a bot of Woolf’s thought process, but when you press go she doesn’t pass the Turing test. I know that there is value to this work—and as I become more comfortable and perhaps apply these tools to non-fiction work as well I can better manage the dissonance that flutters about this exercise.

BONUS:

If you haven’t read Mrs. Dalloway, setting the Voyant Terms Berry to the smallest word sample gives a pretty good summary: 

What is Omikuji?

If you visit shrines and temples in Asia, you may often see people praying for good wishes and taking Omikuji (fortunes written on paper strips) from boxes or even coin-slot machines. The Omikuji predicts your fortunes in health, work, study, marriage, etc. There are many kinds of words written on Omikuji to describe fortunes, and I am interested in the method of using classical Japanese poetry (waka) as divination.

I decided to run some fortune-telling poems with Voyant to see the results. The Omikuji strips are usually rolled up and folded; you will need to unroll them to see the result. Before you read the fortune-telling poems, you will see a general indicator to tell you if you are lucky today. Among many categorization methods, the examples I am using are divided as follows,

• Dai-kichi大吉 (excellent luck)
• Kichi吉 (good luck)
• Chu-kichi中吉 (moderate luck)
• Sho-kichi小吉 (a little bit of luck)
• Sue-kichi末吉 (bad luck; good luck yet to come)

Failure?

I retrieved the data from the Omikuji-joshidosya website. It is said that 70% of current Omikuji strips in Japan are made by the Nishoyamada Shrine, where the Organization Joshidosha (Women’s Road Company) locates.

My first attempt was a total disappointment. See the Figure 4,

The high-frequency words that appeared in Cirrus, TermsBerry, and Trends are single hiragana characters instead of objects’ names and verbs. These words are similar to determiners and prepositions (stopwords in Voyant) in English (the, a, in, to, form, etc.). I then also realized that stopwords are not the only problem in analyzing Japanese text. Text segmentation is also different in Japanese: this issue is already super complicated in modern Japanese, not to mention that the poems in my mini-project are written in classical Japanese. So I tried to refer to the article “Basic Python for Japanese Studies: Using fugashi for Text Segmentation” and see if I could reframe the textual structure of my text for Voyant. For example, I could clean my text before uploading it to Voyant by removing auxiliary verbs, particles, suffixes, prefixes, etc. I also learned about a more manageable solution about Japanese version of stopwords from Japanese DH scholar Nagasaki Kiyonori in his post.

Inspired by Nagasaki Kiyonori, I started to create a stopword list by myself. (Figure 5) The default setting of the stopword list in Voyant Japanese mode is based on modern Japanese. See some examples here,

あそこ　あの　あのかた　あの人　あります　あれ　います　え　おります　から

何　彼　彼女　我々　私　私達　貴方　貴方方

Unlike modern Japanese or Japanese in the Meiji period (1868–1912), auxiliary verbs and particles are almost used in a completely different system in classical Japanese. See some examples in my stopword list here,

が　て　して　で　つつ つ ぬ　たり　り　き　けり　む

But I am glad I chose poetry to do the Voyant experiment because the waka poetry has a relatively easier text segmentation method: one poem always breaks into phrases of 5/7/5/7/7.

Example: 朝日かげ　たゞさす庭の　松が枝に　千代よぶ鶴の　こえののどけさ

Asahikage (5)   tadasasuniwano (7)     matsugaeni (5)   chiyoyobutsuruno (7)     koenonodokesa (7)

Research Questions and Result

Okay, now we have a feasible approach! The next question is about the purpose of this analysis. Should I do a full-text analysis, or should I do several studies with questions that could be asked about those poems? For example, what seasons and figurative language are chosen for good luck and bad luck respectively?

I decided to do a comparison of imagery/actions used in the excellent luck group and the bad luck group. See the number of poems in each group:

• Dai-kichi大吉 (excellent luck) 17
• Kichi吉 (good luck) 6
• Chu-kichi中吉 (moderate luck) 7
• Sho-kichi小吉 (a little bit of luck) 9
• Sue-kichi末吉 (bad luck; good luck yet to come) 11

The result of Dai-kichi大吉 (excellent luck)

Keywords: Spring, breeze, sakura, shadow, rain, sunny, garden, peacefully

The result of Sue-kichi 末吉 (bad luck)

Keywords: autumn, quiet, see, moon, shadow, flower, scatter, reality, top of a tree

The keywords mentioned above have already shown us a sharp comparison between what the creators believe as good luck and bad luck. I am very satisfied with the result, even though I know there are a lot to be improved. I also went to try TermsBerry and Trends in Voyant and realized that I can do further studies using these features. For example although the keyword “flower” and “shadow” both appear in two groups, what associations they have that make them different in good and back luck groups? The example in Figure 8 shows a clear association between flower, sakura (both in hiragana and kanji characters), and peach flower,

Conclusion

The Getting Started with Text Mining is very helpful. I started my mini-project without big data but with the idea that I need to prepare my data (cleaning and removing). If I want to use Voyant to do deeper and larger scale analysis of poems and classical Japanese texts, it definitely requires a huge preparation work. For example, I believe if I do more stopwords considering conjugations, the result probably will be more accurate. I think this tool is great for learning intertextuality and imagery in poetry writing.

There are also Sinitic poetry (kanshi 漢詩) fortune-telling Omikuji! Oh, that would be in a totally different linguist structure, but worth a try next time.

Throughout this week’s readings, I noticed the separation between those who analyze and those who provide the “content” that is analyzed, i.e., the separation between the distant readers/researchers and the authors.

Shall these two never meet?

As a person who writes, I began to wonder whether distant reading my own novel draft might yield some productive insight.

While writing, I often find myself in forest-for-the-trees situations, meaning I am deeply in the mud of the moment of my creation (the frog’s perspective) and feel like I am losing my grasp on the story’s overall arc (the bird’s perspective).

To stay connected to the bird’s eye view, authors who work on longer creative projects (and I suppose longer academic projects, too) will often have either a pinboard with index cards or a writing program like Scrivener with features showing the spine of a story digitally.

However, both of these approaches (index card and/or writing software) are still tied to chronology. And one of the intriguing aspects of distant reading is its promise of simultaneity, of translating a time-based piece into a single image. (Or if not entirely ditching chronology, distant reading at least speeds things up.)

How much of a literary work’s overall concept trickles down/is visible in its more fundamental building blocks (words and sentences)? This is a point of curiosity, a question I had not considered before familiarizing myself with distant reading.

I decided to use Voyant and Word Tree to learn a bit more about my own text-in-progress and see how its micro and macro aspects inform each other.

I uploaded the first chapter of a novel to Voyant I am working on and tried to see whether anything interesting would emerge in the “reveal”. I did not have any specific questions, only many vague curiosities. My belief (based on various experiments conducted for Praxis Assignments throughout the semester) remains that having good initial questions is necessary to find a way for these tools to serve us well. Hopefully, the questions get refined in the process of working with the tools, but an initial curiosity is productive and propulsive.

Here is what I learned about my text:

Most insightful was the Mandala feature in Voyant:
It centered the chapter title “The Idea” and showed a list of salient/defining terms within the chapter. The resulting diagram gave me a snapshot of the chapter as a network. It was satisfying to see the story’s main ingredients, almost as if someone had reverse-engineered the text and created the beginnings of a recipe.

Via the Cirrus feature, I learned that I had certainly established my protagonists/main players in the first chapter. Via Trends I saw that the arcs of the main players intersected in ways that confirmed my intentions and intuitions.

So far, I had received mostly confirmations.

More interesting insight came from looking at the “second tier” of usages, the second largest words in the cloud. I noticed that the program treats the possessive of a noun as a distinct entity. I.e., “Paul” and “Paul’s” are different entities as far as Voyant is concerned. Considering both forms together influences (and in this case amplifies) the overall presence of Paul — which aligns with my intentions but was more difficult to see. The strong presence of “Paul’s” also says something about Paul that I hadn’t explicitly considered: He owns more than others. (More characteristics or more goods? Tbd.)
I can see fruitful research questions emerging around the use of the possessive form in my text and the texts of others.

Another aspect that surprised me was the frequent presence of the word “like”. I would not have anticipated this. Here, Word Tree provided an opportunity to look at the context of these “likes” in more detail.

Based on the specific usages, which I could easily surveil via the word tree above, I might consider stylistic changes. Or perhaps I might notice that simile carries outsized responsibility in my text. The frequency of “like” might point to a theme I could make more explicit in revision. (I am thinking about this.)

In summary:

I can see Voyant being especially helpful in the later stages of the revision process for a novel & when evaluating and implementing editorial feedback. Even when using tools like Voyant on your own writing, the insights distant reading affords are most useful paired with close reading. The data visualization can be an impetus for returning to specific sections and close-reading those. (See also the Richard Jean So and Edwin Roland text “Race and Distant Reading” which details a constructive relationship between close and distant reading by looking more closely at Balwin’s Novel “Giovanni’s Room”)

And one sidebar Q:

I am wondering how Voyant’s Cirrus chooses colors. My text is very much about gender, and I noticed that nouns I had designated male kept coming up as blue and green, female coded nouns as pink. Hmmm. Coincidence? This observation made me want to try the software with a text in a very gendered language (like German).

For my text analysis assignment, I decided to use Voyant to look at one of the English language’s great historical works: The History of the Decline and Fall of the Roman Empire by Edward Gibbon, published in six volumes (further divided into 71 chapters) over thirteen years from 1776 to 1789. Gibbon’s magisterial study spans a period of over 1,300 years, examining the Roman-Mediterranean world from the height of the classical empire to the fall of Constantinople, capital of the eastern Roman empire (which western authors have anachronistically called the ‘Byzantine’ empire since early modernity), to the Ottoman armies of Mehmet II in 1453. Gibbon’s scholarly rigor, sense of historical continuity, and dispassionate, meticulous examination of original, extant sources contributed to the development of the historical method in western scholarship. Nevertheless, some of Gibbon’s conclusions in the Decline and Fall are also a product of the eighteenth century in which the author lived and wrote, and Gibbon’s writing is occasionally punctuated with moralizing statements (briefly touched on below).

The majority of text analysis experiments seem to focus on works of fiction, paying particular attention to their stylistic and aesthetic dimensions. I asked myself: what about an historical work, which is also a narrative construction? Would running Decline and Fall through Voyant allow a reader to observe trends in the stylistic or moralizing dimensions of Gibbon’s grand historical narrative, beyond what might already be grasped by an ordinary reading the text? (disclosure: I have by no means read all six volumes of Decline and Fall in their entirety). 

I used a plaintext file of Decline and Fall from Project Gutenberg that features an 1836-45 revised American edition containing the text of all six volumes. Prior to uploading this file in Voyant, I removed the hundreds of instances of the word ‘return’ in parentheses (which in the HTML version of the file link to the paragraph location in the text), in addition to the preface and legal note at the end of the work authored by Project Gutenberg. After uploading the file, I added additional stopwords to Voyant’s auto-detected list. The terms I removed relate to chapter headings (e.g., Roman numerals), citations (‘tom’ for tome, ‘orat’ for oration, ‘epist’ for epistle, ‘hist’ for historia’, ‘edit’ for edition and so on), and occasional Latinate (‘ad’, ‘et’) and French (‘sur, ‘des’) words. To this end, the word cloud tool was helpful for identifying terms that should be added to the stop-word list.

The resulting word cloud was, to say the least, neither surprising nor particularly revealing nor useful, most of the terms referring to the work’s predominant subjects: 

Standing out as one of the only abstract terms visible in the cloud limited to the top 75 words, however, was “character,” with 828 occurrences. Navigating to the ‘Contexts’ tab, I generated a concordance displaying instances of ‘character,’ which revealed a plethora of specific adjectives used in Gibbon’s text, for example, “manly.” Running ‘manly’ through the ‘links’ generator reveals a network of terms that reflect classical Roman definitions of masculine virtue (‘virtue,’ ‘spirit,’ ‘resolution,’ ‘freedom’) and one usage related to physical appearance (‘countenance’):

These results are once again neither surprising nor interesting, since the ancient (male) writers informing Gibbon’s work were themselves concerned with writing about the meritorious qualities and/or vices of individual male leaders for moralizing, didactic purposes, be they emperors, generals or bishops. This calls to mind Michael Witmore’s observation that “what makes a text a text–its susceptibility to varying levels of address–is a feature of book culture and the flexibility of the textual imagination” (2012). These particular examples may demonstrate the influence of classical authors on Gibbon’s narrative, but they do not necessarily convey anything about what is original in Gibbon’s prose (or, differently stated, original to Gibbon’s contemporary setting).

Perhaps one could get closer to an analysis that better reflects Gibbon’s original, polemical thesis and writing style by first 1) identifying a comprehensive list of moralizing terms (including adjectives like ‘superstitious’ and its variants) harvested from the whole text, and then 2) analyzing the occurrences of those terms in the text, and 3) looking for trends in how those terms are employed throughout the text to describe different social, ethnic, religious or occupational groups. As an enlightenment scholar critical of organized religion, Gibbon maintained that the rise of Christianity led to the fall of the western empire by making its denizens less interested in the present life, including in things civic, commercial, and military, the latter of which would have obvious consequences for the defense of the empire against invasion (Gibbon’s thesis is not generally shared by scholars today). Would such an experiment reveal more exactingly where Gibbon’s moralizing emphases change, based on the chapter or volume of the text where such terms occur?

“What is Distant Reading?”, the title of a NY Times article by Kathryn Schulz has provided one of the simplest ways to understand the topic, “understanding literature not by studying particular texts, but by aggregating and analyzing massive amounts of data.” One might be wondering how to utilize distant reading. In this praxis assignment, I have used Topic modeling, a distant reading approach to analyze historical newspapers.

Newspapers that have survived the course of time are among the most valuable sources of information accessible to academics for researching civilizations and cultures of the past, especially in the context of historical research. Virtually all significant dialogues and disputes in society and culture throughout history were brought to light in newspapers. This is due to the fact that,  as early as the mid-nineteenth century, almost every town regardless of its size saw the establishment of a minimum of one newspaper. Within the newspaper, every facet of the social and daily life of the people is covered with articles such as regarding political debate, promotion of goods and services, and so on and so forth. To this date, no other repository was found with scholarly editorials from history covering controversial topics like political issues, marketing promotion for fashionable clothing, news on major sporting events, and poetry by a local poet all in one place. In a nutshell, for contemporary historians, newspapers record the entire spectrum of human experience better than any other source, giving them access to the past, unlike any other medium.

However, despite their importance, newspapers have remained one of the most underused historical resources for a long time. Historians have found it a daunting task and not to mention sometimes impossible to study historical newspapers page by page for a specific research topic due to the enormous amount and range of material they provide. For instance, just to address one research topic, a historian might have to go through hundreds of thousands of newspaper articles. Ironically, after all these efforts there is still no guarantee of finding the required information.

In this praxis, an attempt will be made to uncover the most interesting and potentially important topics from a period of time using topic modeling on the Paper Past database from DIgitalNZ. As opposed to utilizing the knowledge of historians at the very beginning to find out relevant topics by going through an abundance of newspapers, this praxis relies on a reverse approach. A computational approach will be used for clustering data into topics which will then be evaluated from a historian’s point of view to identify specific patterns in the dataset.

The Experiment

Dataset

Paper Past DigitalNZ is comprised of the following four sections:

1. Newspapers

This section contains digitized newspaper issues from the eighteenth, nineteenth, and twentieth centuries from the New Zealand and Pacific regions. Each newspaper has a page dedicated to providing details about the publication, such as the period range in which it was accessible online. Also, there is an  Explore all newspapers page in which one can discover the URL of all the newspapers. Papers Past contains only a small sample of New Zealand’s total newspaper output during the period covered by the site. But it is more than sufficient for the intended term paper.

During the year 2015, the National Library of New Zealand incorporated a compilation of historical newspapers into their collection that was predominantly targeted at a Mori audience during 1842 and 1935. For this task to be carried out, the University of Waikato’s Computer Science Department used the digital Niupepa Archive, which was created and made accessible by the New Zealand Digital Library Project in 2000.

2. Magazines and Journals
3. Letters and Diaries
4. Parliamentary Papers

Newspaper articles are the particular topic of interest for this praxis. More specifically, the praxis will build a topic model with newspaper articles ranging from 1830-1845. This timeframe is selected because New Zealand announced its declaration of independence in 1835 and this praxis is particularly targeted to find out the topics that emerged in the society during the pre-independence and post-independence declaration period. Paper past provides an excellent API that is open to all. I have gathered 12596 newspaper articles available in the paper past database using the API. The data was migrated into the pandas data frame for further processing and building a topic model on top of it.

I will not be going to discuss the nitty gritty of model building and technical stuff in this article. Instead, I will focus on evaluation and discussion.

Topic Visualization

The visualization is interactive. If you want to check out the visualization, please follow the URL below.

https://zicoabhidey.github.io/pyldavis-topic-modeling-visualization#topic=0&lambda=1&term=

Evaluation and Discussion

The evaluation of the topic model results in this praxis has been done through online browsing and looking for the history of New Zealand during the time period of 1830-1845 along with using general intuition. A historian with special knowledge of New Zealand’s history might have judged better. Some of the topic groups emerged from the topic model results along explanation provided in table 1 for gensim LDA model and table 2 for gensim mallet model. 

Topics	Explanation
“gun”, “heavy”, “colonial_secretary”, “news”, “urge”, “tax”, “thank”, “mail”, “night”‘	Implying political movement and communication during the pre-independence declaration period.
“bill”, “payment”, “say”, “issue”, “sum”, “notice”, “pay”, “deed”, “amount”, “person”	Business-related affairs after the independence declaration.
“distance”, “iron”, “firm”, “dress”, “black”, “mill”, “cloth”, “box”, “wool”, “bar”	Representing industrial affairs mostly related to garments.
“Vessel”, “day”, “take”, “place”, “leave”, “fire”, “ship”, “native”, “water”, “captain”	Represent maritime activities or war from a port city like Wellington.
“land”, “acre”, “company”, “town”, “sale” , “road”, “country”, “plan”, “district”, “section”	Representing real-estate-related activities.
“year”, “make”, “receive”, “take”, “last”, “state”, “new”, “colony”, “great”, “give”	No clear association.
“sail”, “master”, “day”, “passage”, “auckland”, “port”, “brig”, “passenger”, “agent”, “freight”	Representing shipping activities related to Auckland port.
“Say”, “go”, “court”, “take”, “kill”, “prisoner”, “try”, “come”, “witness”, “give”	Representing judicial activities and crime news.
“boy”, “pull”, “flag_staff”, “mount_albert”, “white_pendant”, “descriptive_signal”, “lip”, “battle”, “bride”, “signals_use”	Representing traditional stories about Maori Myth and Legend regarding mount Albert.

Table 1: some of the topics and explanations from gensim LDA model

Topics	Explanation
‘land’, ‘company’, ‘purchase’, ‘colony’, ‘claim’, ‘price’, ‘acre’, ‘make’, ‘system’, ‘title’	Representing real-estate-related activities.
‘native’, ‘man’, ‘fire’, ‘captain’, ‘leave’, ‘place’, ‘officer’, ‘arrive’, ‘chief’,  ‘make’	Representing news regarding New Zealand War.
‘government’, ‘native’, ‘country’, ‘settler’, ‘colony’, ‘man’, ‘act’, ‘people’, ‘law’	Representing news about the sovereignty treaty signed in 1835.
‘mile’, ‘water’, ‘river’, ‘vessel’, ‘foot’, ‘island’, ‘native’, ‘side’, ‘boat’, ‘harbour’	Representing maritime activities from a port city like Wellington.
‘settlement’, ‘company’, ‘make’,’war’, ‘place’, ‘port_nicholson’, ‘settler’, ‘state’, ‘colonist’, ‘colony’	Representing news about Port Nicholson during the war in Wellington 1839

Table 2: some of the topics and explanations from gensim Mallet model

Since starting my academic journey in DH over two years ago, I’ve been awaiting the moment when I’ll get to learn about text mining/analysis tools. I’ve worked in the “content” space my entire career and I’ve always been interested in the myriad tools out there that allow for new ways to look at the written word. I spent nearly a decade as an editor in the publishing world, and I never leveraged an actual text analysis tool, but I jerry-rigged my own approach for scouring the web for proper usage when I found myself confused on how to best render a phrase. My go-to text analysis “hack” has always been to search google for a phrase I’m unsure of in quotes, and then to add “nytimes.com” to my search query. This is based on my trust that the copyeditors at NYT are top notch and whatever usage they use most often is likely the correct usage. For instance, if I encounter the usage of “effect change” in some copy I’m editing and I’m not sure whether it should be “affect change,” I would do two separate searches in Google.

1. “affect change” nytimes.com
2. “effect change” nytimes.com

The first search result comes up with 72,000 results. The second result comes up with 412,000 results. Thus, after combing through the way the term is used in some of the top results, I can confidently STET the use of effect change and move on without worrying that I’ve let an error fly in the text. I’ve used this trick for years and it’s served me well, and it’s really as far as my experiments in text mining have gone until this Praxis assignment.

Diving into this Praxis assignment, I was immediately excited to see the Google NGram Viewer. I had never heard of this tool despite working in a fairly adjacent space for years. Obviously, the most exciting aspect of this tool is its absolute ease of use. It runs on simple Boolean logic and spits out digestible data visualizations immediately. I decided to test it out by using some “new” words to see how they’ve gained in published usage over the years. I follow the OED on Twitter and recall their annual new words list announcement, which for 2022 was produced as a blog post doing its best to leverage the newest additions in its text. You can read the post here: https://public.oed.com/blog/oed-september-2022-release-notes-new-words/

The NGram has a maximum number of terms you can input, so I chose the words and phrases that jumped out at me as most interesting.

The words I chose from the post were (in order of their recent frequency as spit out by NGram): jabbed, influencer, energy poverty, side hustle, top banana, Damfino, mandem, and medical indigency. As you can see, most of these terms are all quite new to the published lexicon — all but “jabbed.” However, jabbed in the early 20th century likely had more to do with boxing literature than vaccinations.

Moving along in this vein, I then looked up the “word of the year” winners dating back the last decade. These words were: omnishambles, selfie, vape, post-truth, youthquake, toxic, climate emergency, and vax. 2020 did not have a word of the year for reasons I suspect have to do with the global pandemic. Looking at the prominence of these words in published literature over the years showed a fairly similar result as the “new” words list.

What I found surprising is that these words and phrases are actually “newer” than the ones I pulled from the new words list. There’s barely a ripple for all these words outside of “toxic,” which has held populat usage for over a century now according to NGram.

Despite to say, as a person who routinely looks up usages for professional purposes, I’m elated to discover this tool. It will not only help me in my DH studies, but will also assist me in editorial work as I look for the more popular usage of terms. Instead of having to us Google’s own search engine and discern the results myself, I can now see simple visualizations that will prove one usage’s prominence over another.

NGram is well and good, but I could tell this was a bit of a cop out when it came to learning the ins and outs of text mining. So I decided to test out Voyant Tools to see if I could get a handle on that. As was noted in the documentation, it is best to use a text I am familiar with so I can make some qualitative observations on the data this is spit out. I decided to use my recently submitted thesis in educational psychology, as there’s likely not much else I’m more familiar with. My thesis is titled, “User Experience Content Strategies for Mitigating the Digital Divide on Digitally Based Assessments.” Voyant spat out a word cloud that basically spelled out my title in via word vomit in a pretty gratifying manner.

This honestly would have been a wonderful tool to leverage for the thesis itself. As I tested 200 students on their ability to identify what certain accessibility tools offered on digital exams do, I had a ton of written data from these students and I could have created some highly interesting visualizations of all the different descriptive words the students used when trying to describe what a screen reader button does.

I’ve always known that text analysis tools existed and were somewhat at my disposal, yet I’ve never even ventured to read about them until this assignment. I’m surprised by how easy they are to get started with and am excited about leveraging more throughout my DH studies.

For my text mining praxis assignment, I decided to use Python’s Natural Language Processing (NLP) package, also called Natural Language Toolkit (NLTK). Further to last month’s text analysis workshop, I thought it would be a good idea to put into practice what learnt.

I picked Jane Eyre, a book I read few times in both the original language and a couple of translations, to ensure I could review the results with a critical eye. The idea was to utilise NLP tools to get an understanding of the sentiment of the book and a summary of its contents.

In an attempt to practice as much as possible, I opted for an exploratory approach to trial as different features and libraries. At the bottom of this post, you will find my Jupyter notebook (uploaded on Google Colab) – please note that some of the outputs exceed the output space limit, hence you might need to open them in your editor. In terms of steps undertaken, I was able to create a way to “assess” words (positive, negative, neutral) and run a cumulative analysis on big part of texts to gather a sense on how the story develops. Separately, I wrote a function to summarise the book, which drove the length of it from 351 pages to 217 (preface, credits, and references included), however I am not sure about the quality of the result!! Here in PDF this “magic” summary!

Clearly the title of my blog post is meant to be provocative, but I can see how these algorithms could be used as shortcuts!

Before diving into the notebook though, I would like to share a few of thoughts on my takeaways on text analysis. On the one hand, it is impressive to witness the extent to which the latest technologies can take unstructured data, interpret it, translate it into a machine-readable format, and then make it available to the user for further analysis or manipulation. Machines can now operate information extraction (IE) tasks by mining meaningful pieces of information from texts with the aim of identifying specific data, or targets, in contexts where the language used is natural and the starting dataset is either semi or fully unstructured. On the other hand, I personally have concerns around the fact that text mining softwares perform semantic analyses that often can only leverage a subsection of the broader spectrum of knowledge. This is to say that the results produced by these technologies can certainly be valid, however, could be limited by the inputs and related pre-coded semantics, hence potentially translating into ambiguous outputs.

There is a chance that the below HTML will not render the notebook, if so, you can download it directly from myGitHubGist.

A/N: This post contains a lot of information about my project, Map Lemon. If you don’t want to be deeply confused about what Map Lemon is and why it is, you can head on over to my blog at https://noveldrawl.commons.gc.cuny.edu/research/, as I’m not explaining it for the sake of brevity in this post. The corpus itself is not yet publicly available, so you’ll just have to trust me on the docs I’m using for now.

I’ll start this post with a bit of introduction on my background with text mining. I’m a Linguist. I’m a Computational Linguist. And more importantly than either of those two really nebulous phrases that I’m still convinced don’t mean much, I made a corpus. I am making a corpus, rather (everything is constantly moving changing and growing). It happened by accident, out of necessity—it wasn’t my original research interest but now that I’m deep in it I love it.

My corpus, Map Lemon, is #NotLikeOtherCorpuses (I’m sorry for that joke). It’s not text mined. A LOT of linguistic corpuses are text mined these days and that gets on my nerves in a real bad way. Here’s why:

Let’s use the example that you’re text mining on Twitter to get a better idea of jargon used within a niche community, since this use-case is quite common.

1. Text mining often takes phrases out of their contexts because of the way platforms like Twitter are structured.
2. These aren’t phrases that, generally speaking, are used in natural speech or writing. While cataloging internet speak is important, especially to understand how it affects natural S&W, we’re not cataloging as much natural S&W as a result, and I don’t think I need to explain why that’s important.
3. It’s not situational. You’re not going to find recipes for lemonade, or directions to a lemonade stand yes I’m making a joke about my own research here on Twitter.
4. You’re often missing demographics that are unknown that can affect the content of the corpus.

I chose to do this praxis assignment to debunk, or at least attempt to, all of those things. I want text mining to work for me. It probably won’t for my use-case, but I should at least be versed in doing it!

Now let’s get into it.

I decided to text mine my own corpus. Yup. I’m at a stand-still with the results I’ve been getting and need material for an abstract submission. Here we go.

So, since my data has already been cleaned before, I went ahead and just plopped it into Voyant. The following ensued:

• Oh, rats I forgot I can’t just do that with all the demographics and stuff in there and confuse Voyant.
• Okay, copy the file. Take out all the not responses. Might as well separate them into their respective experiments while I’m at it.

So, the final version of what I’m analyzing is: 1) Just the directions to the lemonade stand 2) Just the recipes for lemonade. I’m not analyzing the entire corpus together since it wouldn’t yield coherent results for this specific purpose due to the difference in terminology used for these two tasks and lack of context for that terminology.

The results from the directions were really neat in that you can follow the correlations and word counts as directions given, basically. Here’s the map from Experiment I so you can follow along:

Here are the most common phrases in the directions given, according to analysis with Voyant:

• “before the water fountain”; count: 4
• “take the first left”; count: 4
• “a carousel on your left”; count: 3
• Some other phrases that are all count 3 and not as interesting until…
• “at the water fountain”; count: 3
• “between the carousel and the pond”; count: 3

Now, okay, these numbers aren’t impressive at first glance. Map Lemon only has 185 responses at present, so numbers like this maybe aren’t all that significant, but they sure are interesting. Map Lemon contains exclusively responses from North Americans, so from this we could postulate that North Americans tend to call “that thing over yonder” a water fountain or a carousel. But also from this we can see the directions Chad gets most commonly: people often send him down the first left on the street; of the group that does not, and has him cut through the park, they let him know that he should pass the carousel on the left; and the lemonade stand is just before the water fountain. All these directions are reiterated in two different ways, so it seems. That sure is neat! Not particularly helpful, but neat.

So let’s look at those cool correlations I mentioned.

• ‘gym’ and ‘jungle’ – correlation: 1 (strongest)
• ‘clearing’ and ‘paved’ – correlation: 1
• This one I’m unsure what is really meant by it, if that makes sense, but it was ‘enter’ and ‘fork’ corr. 1
• ‘home’ and ‘passed’ – correlation: 1

These look like directions alright! Okay, of course there’s the phrase ‘jungle gym’, but we do see, okay, there’s a paved clearing. I’m sure at some point Chad has to enter a fork, although I’m a bit confused by that result, and yes, many people did have Chad pass the house. Neat!

I’m a bit skeptical of some of these correlations as well, because it’s correlating words strongly that only appear once, and that’s just not a helpful strong correlation. But that’s just how the tool works.

Looking at contexts wasn’t particularly helpful for the directions, as lot of the contexts were for the words ‘right’ and ‘left’.

Now, here’s what was really freakin’ cool: the links. Voyant made this cool lil graphic where I can see all the most common words and their links. And it actually shows… directions! The 2/3 most common paths, all right there, distilled down. Try giving Chad directions for yourself and see what I mean, ‘cause it’ll probably look something like this:

Voyant’s link map for Map Lemon Experiment I

Okay, so the directions didn’t show anything revolutionary, but it was pretty darn cool. Let’s move onto the recipe analysis.

NOW THIS IS FREAKIN’ COOL!!! According to the phrase count tool, everybody makes lemonade about the same! Including using a lot of the same amounts of ingredients and even the same filler phrases!

Ingredients:

• 1 cup sugar; count: 3 (the semantics of this compared to the other two is really interesting!)
• 3 cups of water; count: 3
• 4 lemons; count: 3

Filler phrases:

• “a lot of”; count: 5
• “make sure you have”; count: 5
• “kind of”; count: 4 (context for this one is tricky)

Perhaps that’s a recipe I could try!

Now, okay. If we’re skinning cats, there’s not a lot of ways to skin this one, actually. We specifically chose lemonade for this experiment because it’s ubiquitous in North America and really easy. And the correlations feature wasn’t really helpful this time around for that exact reason (lack of distinguishing words). But look at this cool link map thingy!!

Voyant’s link map for Experiment II

Very helpful! You either squeeze, cut, juice, or halve your lemons—not all four (and two of those are only different out of semantics). Add several cups of water to a pitcher, and stir in (or add—again, semantics) at least a cup of sugar. Boom! There’s our lemonade recipe. This was so cool to synthesize!

At the end of this little project, I am still as annoyed with the inability to use demographics with text mining as I was before. This means that demographics for text mining for my purposes would have to be very carefully limited to control for individual linguistic variants. However, I also see the benefit in this tech; although, I definitely think it’d be better for either larger datasets or datasets that have a very controlled demographic (so that small numbers like 3 and 4 are actually statistically significant). Mostly, for now, it seems that the results just make me say “cool!”. I love the visual links a lot; it’s a really good example of graphics that are both informative and useful. I think it would be a fun side project to try and synthesize a true “All-American Man” using text mining like this. (Side note, that exact sentence reminds me that in my Computational Linguistics class in undergrad, for our final project about half the class teamed up and scraped our professor’s Twitter account and made a bot that wrote Tweets like he did. It actually wasn’t all that bad!)

I think this could potentially be valuable in future applications of my research, but again, I think I need to really narrow down the demographics and amount of data I feed Voyant. I’m going to keep working with this, and maybe also try Google Ngram Viewer.