2

Narrative Research in the Social and Behavioral Sciences

The first session of the workshop was designed to set the stage for the day by providing an overview of research related to narratives in the social sciences and humanities, explained moderator Jeffrey Johnson, University of Florida. He noted that the four panelists had been asked to discuss the state of the art in this research from the perspectives of diverse fields, and to explore ways in which new technologies, particularly big data (very large datasets that can be mined or analyzed using only computer technology), can expand the possibilities for analysis of narrative.

IDENTIFYING NARRATIVE STRUCTURE WITH QUANTITATIVE METHODS

Although the use of quantitative methods in narrative research is common practice today, this was not always the case, according to Roberto Franzosi, Emory University. A pioneer in quantitative narrative analysis, Franzosi explained that he first became interested in narrative structure while using statistical models to study time series data on the labor strikes of post–World War I Italy. At the conclusion of his research, however, he concluded that these models were limited in their ability to identify the social actors that were at the heart of these strikes. To address this limitation, he began to use newspapers as a new data source for identifying social actors. While he was conducting his analysis, it became clear to him that all narratives have an invariant structure, which includes a subject (social actor), a verb (social action), and an object. According to Franzosi, this structure (SVO) is rooted in rhetoric (the study of persuasive communica-

tion), which he explained was first introduced in ancient Greece, and later translated by Thomas Wilson in The Arte of Rhetorique¹ to what are now often referred to as the five W’s of journalism (who, what, where, when, and why).

Franzosi pointed out that because of the immense volume of data available today, computational methods are a necessity in narrative research. Therefore, he developed software that would allow him to identify the SVO structure in narratives quickly and accurately using a method he terms “quantitative narrative analysis.” With this software, Program for the Computer-Assisted Coding of Events, he was able to examine approximately 50,000 newspaper articles from 1919 to 1922, which resulted in the identification of 250,000 SVO sequence sets related to the rise of Italian fascism in postwar Italy.

Returning to the five W’s structure, Franzosi observed that while the SVO structure addresses the questions who and what, geographic information system (GIS) models are helpful in identifying where actors act. As an example, he referred to his research on the emergence of Italian fascism, in which he used a GIS model to track the activity of socialists and fascists from 1919 to 1922. He found that the location of socialists leading the revolutionist movement from 1919 to 1920 overlapped with the locations in which fascism first emerged in 1921 and 1922.

Franzosi closed by sharing two different approaches currently being used by researchers to extract the SVO structure in narrative analysis. He first described ClausIE,² German open-source freeware software based on the Stanford CoreNLP natural language software.³ This online open information extractor uses automated methods to identify the SVO structure in text. Franzosi characterized the second approach as similar to novelist Kurt Vonnegut’s “man in a hole” narrative. Vonnegut posited, Franzosi explained, that stories often follow specific patterns such as “man in a hole.” In this narrative, a character who is relatively happy receives bad news (or something bad happens to him). This bad news causes a temporary dip in the man’s state of mind or welfare until his luck inevitably changes, and he receives good news (or something good happens to him), which pulls him out of the “hole.” According to Franzosi, researchers are now able to plot these arcs in narrative by using sentiment analysis (computational analysis used to identify the intended sentiment for a set of words) and complex

___________________

¹ Wilson, T. (2010). The Arte of Rhetorique (1560). Oxford, UK: Benediction Classics.

² For more information on ClausIE, see https://www.mpi-inf.mpg.de/departments/databases-and-information-systems/software/clausie [April 2018].

³ For more information on Stanford CoreNLP, see https://stanfordnlp.github.io/CoreNLP [April 2018].

algebraic matrix factorizations, such as singular-value decomposition and non-negative matrix factorization.

ADVANCING NARRATIVE RESEARCH WITH BIG DATA SCIENCE

Mark Turner, Case Western Reserve University, first conceived the idea of building a data repository for the study of communication in the early 2000s. He had observed large laboratories in the genomic and other biological sciences working together to build large online databases to store and share data collectively, and said he hoped to provide a similar resource for studying different modes of communication. In 2010, he and a colleague developed Red Hen Lab, a global consortium for communication research.⁴ According to Turner, the repository currently stores approximately 4 billion words and 360,000 recorded hours of audiovisual (AV) broadcasts that are searchable using natural language processing (NLP) tagging tools, such as Stanford CoreNLP and Apache OpenNLP, and optical character recognition software, such as Tesseract. The most common type of data contained in the repository, he noted, is AV news broadcasts in such languages as English, German, Portuguese, Russian, Czech, Arabic, and Chinese.

Turner explained that, although the main goal of the consortium is the development of theory, he and his colleagues are also interested in developing the computational methods needed to conduct theory-based research. According to Turner, for example, the repository currently has the ability to identify and tag frames—cognitive structures that determine the process and result of interpreting linguistic forms, such as words, phrases, and grammatical patterns⁵—so that they can be searched along with grammar and construction and word strings. To illustrate the use of this method, he cited its ability to help researchers identify locations where there is an emerging risk to a region’s stability because of such significant events as extreme weather. He added that the consortium also benefits from the expertise of researchers and students in other disciplines, including linguistics and machine learning, who share with the consortium methods and techniques that further advance the study of narrative. Turner closed by noting that these methods, combined with the consortium’s data repository, enable researchers from around the world to track and study narratives in real time.

___________________

⁴ For more information on Red Hen Lab, see http://www.redhenlab.org [April 2018].

⁵ Heine, B., and Narrog, H. (2009). The Oxford Handbook of Linguistic Analysis. Oxford, UK: Oxford University Press.

FUNCTION WORDS AND THE ARC OF NARRATIVE

James Pennebaker, University of Texas, Austin, examines narrative from a psychological perspective. He explained that he was first introduced to narrative research while conducting an experiment on expressive writing. It is well known in medicine and psychology, he observed, that individuals who have experienced a major traumatic event at some point in their life are much more likely to develop health problems, and the likelihood increases if their traumatic experience has been kept secret. Speculating on the effect writing about these secret traumatic experiences might have on a patient’s health, he conducted a study in which participants were randomly assigned to write about either a personal trauma or a less personal, superficial experience. He and his colleagues then monitored the participants’ health using such markers as immune function. The results, he reported, showed that the act of writing about a personal trauma can produce positive changes in physical health.

Pennebaker and his students then developed a software program that would allow them to analyze text more efficiently. This word counting software program, he explained, known as Linguistic Inquiry and Word Count (LIWC), made it possible to analyze text in a new way, providing the ability to calculate the percentage of positive and negative emotional words; cognitive process words, such as “understand,” “realize,” “know,” or “meaning”; and parts of speech, such as pronouns, prepositions, articles, conjunctions, and auxiliary verbs.

Using this computational analysis, Pennebaker learned that the way people tell a story can often be more revealing than the story’s content. By looking at function words (e.g., pronouns, prepositions, articles, conjunctions, auxiliary verbs),⁶ for example, he and his colleagues were able to identify a surprising number of characteristics of the authors of the stories being analyzed, including their genders, intelligence levels, emotional states, and social connections. The use of function words, they found, could reveal whether the author was being honest or not.

Intrigued by these results, Pennebaker conducted additional studies using LIWC analysis. In one study, he analyzed stories created by students after they had participated in a thematic apperception test.⁷ The study revealed, he reported, that graphs showing the number of function words used in a story typically reveal a particular pattern: they are used at a high

___________________

⁶ Pennebaker explained that function words are social words that are processed differently in the brain than are content words, which include nouns, regular verbs, most adjectives, and most adverbs.

⁷ A thematic apperception test is a psychological test used to reveal thoughts and attitude patterns by examining the stories people devise when presented with a picture involving people unfamiliar to them. For more information, see http://www.utpsyc.org/TATintro [April 2018].

rate in the beginning and then eventually drop off, creating an arc. In what became known as the “arc of narrative” project, he analyzed thousands of novels, short stories, Supreme Court decisions, and movie scripts. The arc of narrative pattern, he discovered, can be predictive of positive emotional states and successful outcomes. For example, he and his colleagues analyzed stories that detailed romantic breakups. They found that the arc of narrative pattern was present in such stories told by authors who had recovered from the event but not in those told by authors who had not yet recovered. Similarly, they found that movie scripts containing the arc of narrative pattern had higher ratings than those that did not.

Pennebaker closed by observing that he and his colleagues are currently analyzing stories told by members of ISIS, as well as ones already identified as having been told by liars and truth-tellers. Such research, he explained, is allowing researchers to gain understanding of how people and groups construct their stories and histories.

THE SMALL-STORY APPROACH TO NARRATIVE RESEARCH

Michael Bamberg, Clark University, divides narrative analysis into two categories: big-story and small-story approaches. Big stories, he said, help people make sense of the world through the narratives told by nations, organizations, institutions, and individuals, citing the example of the story of how the allies sought to contain the spread of communism after World War II. He defined small stories, on the other hand, as everyday stories such as fairytales, novels, and personal conversations. Narrative analysis, he explained, typically focuses on the big stories and addresses the structural, textual, and thematic aspects of their narratives, whereas he is more interested in when, why, and how small stories influence big stories.

Bamberg described research he conducted based on the hypothesis that one way small stories differ from big stories is in their formation. He noted that while some small stories, such as novels and fairytales, are presented in familiar structures, conversational small stories are not, and because they are regarded as mundane, they are often overlooked. He explained that his small-story approach to narrative analysis examines how a story originates, is picked up by others, and later transforms over time. He added that he used several methods common in communications research, such as conversation analysis of both verbal and nonverbal communication. He noted that gestures, posture, and facial expressions can often reveal how a story is received and understood by others. He suggested, however, that more research is needed on visual narratives, such as those used in commercials to facilitate highly emotional plot structures.

In closing, Bamberg pointed out that while small stories are often formed and shaped by big stories, it is also likely that big stories form from

small stories. Similarly, he said, it is possible for small stories to challenge or change big stories. He concluded by suggesting that research to improve understanding of these two phenomena would be beneficial for intelligence analysis.

DISCUSSION

Turner opened the discussion by agreeing with Bamberg that research on visual narratives is needed. He noted that he and his colleagues at the consortium have been collecting and tagging a variety of visual data to store in the Red Hen Lab repository, adding that even subtle facial expressions and gestures can be detected and tagged so that researchers have access to all forms of communication.

The remainder of the discussion focused on comparing methodological approaches and considering how narrative research might help analysts determine whether a narrative is true or false and whether it is changing at the societal level.

One participant commented on the range of approaches presented by the speakers. Franzosi and Pennebaker, he suggested, focus their analysis on text and sentence structure, while Bamberg’s approach is more qualitative. The consortium for data science discussed in Turner’s presentation sits somewhere between these two approaches, he added. Pennebaker responded that both qualitative and quantitative approaches are necessary for narrative research. Elaborating, he explained that although the quantitative methods used in his work are effective and accurate, they are complemented by the more individualized qualitative methods of studying narrative. Turner was asked whether it is possible for big data researchers to understand their data as well as qualitative researchers working in the field. Turner asserted that, while qualitative researchers are more familiar with their data, computational and statistical methods may provide insights not available through qualitative methods. He added that members of the consortium plan to continue collecting data that are multimodal and current so analysts can observe narratives as they evolve in real time.

Responding to a similar comment on the importance of measuring outcomes, Pennebaker agreed and suggested the need for a shift from the use of traditional outcome questionnaires to measuring behavior to determine how narrative impacts a person’s life. He commented that the research discussed by Franzosi is important because, by revealing that the manipulation of language can cause changes in a person’s emotional state, it revealed a connection between language and behavior.

Turning to the topic of rhetoric, one participant noted that analysts are often asked to make sense of a situation on the basis of limited information. However, since narratives are designed to be persuasive to both the person

constructing the narrative and others, he wondered whether the result could be analyses that are more imagination than reality. Another participant called attention to the significant body of research in narrative theory on fictionality, which separates the idea of general fiction (e.g., novels, films) from the idea of fictionality as a method of invention.

Noting that analysts are asked to make predictions on both the strategic and tactical levels, one participant asked how a research model might reveal when a phenomenon such as preference falsification (when people are expressing what they consider to be socially acceptable preferences rather than their true preferences) is occurring. If an analyst is aware only of the societal narratives created by preference falsification, she added, it may be more difficult to predict when social change is about to occur. Pennebaker explained that studying the use of function words can reveal things the speaker or author may be trying to hide with the use of content words. For example, he said, by studying George W. Bush’s speeches, he could identify when Bush decided to enter the war in Iraq. Approximately 9 months before doing so, he elaborated, Bush significantly decreased his use of the pronoun “I.” According to Pennebaker, use of the word “I” indicates when someone is being personal. Only after the United States entered Iraq, he added, did Bush return to using the word “I.” Pennebaker also noted that the Boston Marathon bomber stopped using the word “I” in his social media communications before the bombing.

This page intentionally left blank.