This year the team of Impact and Fiction project was represented on one of the most important events in the world of Digital Humanities – ADHO Digital Humanities Conference which took place on-site in a beautiful city of Graz, Austria. It was the first time in the last 4 years that the conference was held with physical attendance because of “digital pandemic years”. Although DH in the abbreviation of the conference title (DH2023) stands for Digital Humanities, humanists still love to have real contact with real people.
The special topic of the conference was “Collaboration and Opportunity” which is very true for the DH field: creative co-operation between historians, art historians, literary critics, linguists on the one hand and engineers, software developers on the other creates new opportunities for research and for its applications. This year’s conference also showcased transdisciplinary and transnational collaboration.
From the Impact&Fiction project there were 3 presentations regarding the measuring the impact of literary texts.
Have you ever read a book which made you cry and just after that on the next page you couldn’t help laughing out loud? Or remember the case of how, reading, you almost burst with anger, but already in the next chapter felt tremendous relief. Sometimes when we read a book, it’s as if we’re on an emotional seesaw. We hypotised that this could be a good measurement of impact and we used the notion of valence after among others Jacobs 2019.
How do you gauge valence of a book? It is not so simple, since firstly the computer can not feel anythink so we have to code its feelings and secondly the feelings of readers can vary. There are researchers though who came up with dictionaries of words carrying sentiment, and we used these dictionaries to check our theory. The challenge was to apply the valence methods to books or at least large samples from books, since before we conducted sentiment and valence analysis mainly on a word-level. There were a few researchers trying to apply the same techniques beyond the word, however the sample was not larger than 100 words at a time.
The challenge was accepted, but alas, the results gave more questions than answers for now. First of all, the theoretical background of the experiments we conducted with Dutch literary texts is not even: there is no single well-defined concept of valence and its definitions vary from the perspective of the author, world and reader. Second, the methods which can work well on a word-level do not give appropriate results since almost all of them assume that the context is irrelevant which is not so when we speak about literary books. Besides, in books we can find many ambivalent passages which are difficult to interpret from the valence point of view. And last but not least, the readers are different in their evaluation of emotional impact and so the perception of (word) valence varys as well.
2. Affective-Aesthetic Potential
If it seems yet too difficult to gauge the valence of the whole book or even the corpus of books, maybe we should try “to eat an elefant by bite at a time” and consider one of the components of valence. We see this component as affective-aesthetic potential (AAP) of words. The idea that words, like chemical elements capable of attaching other elements evoking some reaction, can have aesthetic potential and thus evoke a certain reaction in the reader was first voiced by Jacobs in 2017 and includes the idea that even the sound shell of a word can influence the perception of that word. However, AAP is calculated using the following formula: the sum of words close to words with positive colouring minus the sum of words close to words with negative colouring, with positive and negative words defined in advance. To make it simpel: firstly we have to look how close the words are to the positive group pre-selected words or to the negative group and then just sum up these two groups and subtract one from another.
Where can we find the pre-selected words and if this group of words is big? Good questions. In our study we used the 130 words proposed by Jacobs in his article “Quantifying the beauty of words” (2017). It seems though that there is not really a difference between measuring the AAP of words or the valence of the text, the result stays vague. We even contrasted the AAP scores based on Wikipedia embedding (Y) and the AAP scores based on novel embedding (X):
This “cloud” reflects that there is probably no distinctive difference between the two embeddings. we could speculate a bit saying that the novel embedding leads to a wider range of scores so that we could say that the AAP is more likely to be a feature of a literary text, but ist is not really true. That means that the beautiful idea of words having an affective-aesthetic potential is not proven empirically yet.
3. Perception of style
If we can not yet measure the impact of a literary text by looking at the text itself yet, we could take a look at online book reviews and find if the stylistic features which could affect the readers are mentioned there. Therefore we annotated a small set of randomly chosen 500 reviews from the online Dutch book reviews corpus (ODBR, Boot 2017), tagged the sentences where style was mentioned and categorised the tags. We found a reflection on style in 45% of the annotated reviews. Almost 19% of these refer to specific elements of the text.
Focusing on features of the text we can see that most reviews mention chapters, which is close to the book structure, and the most often mentioned characteristic of chapters is length. However, in terms of stylistics it is more interesting to look at the next frequently mentioned textual elements, such as sentences, words, perspective and time.
Since sentences and words are frequently used to examine the theory of foregrounding from which we proceed in the talk, it is interesting to mention how readers evoluate perspective and time. Most reactions on stylistic realisation of perspective include an indication of which grammatical person the narrative uses: whether the text is written in the first, second or third person. We use the whole corpus to understand exactly how readers react to the use of a particular pronoun in the text. It turnes out that there were far more reactions to texts written in the first person than to texts written in the second or third person. This is true across all genres.
The first person is also often connected with the time of the narrative and therefore has an impact on the readers. For example, someone wrote:
“The story is written from the first person perspective and in the present tense, which increases the reader’s involvement in the event.”
It also turns out that some readers perceive the use of the first person pronouns in the past tense as a challenge:
“The story is written in the first-person form but some people may have to get used to it in the past tense.”
We consider the pronounce-time relation remarkable and requiring deeper study in the context of understanding stylistic complexity.
Can science provide clear answers to the questions put to it? Of course, but in this case such questions are no longer scientific, but rather practical. Once the answer is given, science ends. Because science is a constant search for questions that give rise to new questions. Therefore, we were happy to present our research that refutes some theoretical positions to our colleagues in order to come up with new questions and new ideas about how literature affects us and, most importantly, how we can measure it. There is a long way to go, this blog will keep you up to date with our new steps along the way.