Roberto Franzosi

From Words to Numbers: A Generalized and Linguistics-Based Coding Procedure for Collecting Textual Data

In this paper, I describe a new, powerful technique for coding data from textual sources, a technique based on concepts developed in thefield of linguistics, particularly on the concept of semantic text grammar. I develop a grammar of data collection from the simple linguistic canonical structure noun phraselverb phrase, namely, subjectlactionlobject and their modifiers. I show that the grammar allows researchers to collect richer and more flexible data than more traditional coding schemes. In particular, the grammar produces coded output that, to a large extent, preserves both the syntax and the lexicon of the source material. Furthermore, the grammar makes the process of data collection independent of any prior specification of hypotheses, because virtually all relevant information provided by the sources can be easily coded.

Ways of Measuring Agency An Application of Quantitative Narrative Analysis to Lynchings in Georgia (1875–1930)

This paper advocates an actor-centered, relational view of agency and proposes Quantitative Narrative Analysis (QNA) as a promising method for operationalizing and measuring agency. QNA organizes the information contained in narrative texts by exploiting the invariant linguistic structural properties of narrative—namely, sets of SVOs (Subject, Verb, Object) organized in predictable sequences and where in narrative S are actors and V are actions. The relational data made available by QNA are ideally suited for analysis with geographic information systems (GIS) tools, sequence analysis, or network analysis. These tools preserve the centrality of agency (actors and their actions) in social scientific explanation of social reality. An application of QNA to newspaper stories of lynchings in Georgia (1875–1930) will illustrate the power of this approach. The paper complements the illustration of this quantitative way of measuring agency with discourse analysis—another popular social science approach to texts. We will rely on this approach to illustrate how linguistic and rhetorical strategies can be used to hide agency in texts and the challenges (and solutions) this poses for measurement: How can we measure something that is not there?

New directions in formalization and historical analysis

In the early 1960s, historians and sociologists in American universities began a rather curious dialogue. Individuals from both disciplinary camps had begun making regular forays across the scholarly divide. The translation into English of Max Webers work inspired a number of American sociologists (Reinhard Bendix, Barrington Moore, Neil Smelser, and Charles Tilly, for example) to begin raising historical questions, pursuing them through detailed empirical scholarship and bringing to bear the theoretical sensibilities and formal analytic techniques that characterized mainstream social science. At the same time many historians, among them Lee Benson, Stephen Thernstrom, Robert Fogel, and (at the University of Iowa) Allan Bogue, William Aydelotte, and Samuel Hayes had begun to make use of computer aided quantitative analysis and formal analytic approaches to address traditional historical disputes and also to raise new social scientifically informed questions.1

Network analysis of narrative content in large corpora

We present a methodology for the extraction of narrative information from a large corpus. The key idea is to transform the corpus into a network, formed by linking the key actors and objects of the narration, and then to analyse this network to extract information about their relations. By representing information into a single network it is possible to infer relations between these entities, including when they have never been mentioned together. We discuss various types of information that can be extracted by our method, various ways to validate the information extracted and two different application scenarios. Our methodology is very scalable, and addresses specific research needs in social sciences.

The Difficulty of Mixed-Method Approaches

On July 9–10, 2012, Professor Martin Bauer organized a third conference on text mining at the London School of Economics. Invited participants came from computer science and different social science disciplines. They were united by a common interest in texts and the extraction of meaning from those texts. They were divided by the way they approached those texts—their methods, a marker of more fundamental epistemological differences. Social scientists ranged in their approach. There are the content analysts who rely on CAQDAS packages to deal with text, either qualitatively, employing CAQDAS only as an organizational tool, or quantitatively, querying the codes they produce in CAQDAS for cross-relations, exporting the results, and analyzing these in statistical packages (SPSS, STATA, R), as do Michael White, Maya Judd, and Simone Poliandri (this volume, 2012:43–76). There are those who distance themselves from the text, relying on semi-automated approaches in the French tradition of analyse des données (e.g., using such software as Alceste or Lebart’s Dtm-Vic). The larger the body of text to be analyzed, the more likely social scientists are to look for automatic, quantitative solutions. And that’s where social and computer scientists meet. The textual analysis tools that computer science data miners have developed tend to work well only when dealing with ‘‘millions of words,’’ as Professor Nello Cristianini, a computer scientist at Bristol, UK, put it at the conference: ‘‘Modern AI is driven by large text corpora.’’ Their algorithms do allow them to reveal important properties of content, whether about topic, style, or even narrative structure. It is that narrative structure (the sequence of SVO of actors and their actions), that I painfully extract from text by hand, in my Quantitative Narrative Analysis approach—an approach that falls squarely in the category of those who treat text qualitatively but analyze the results quantitatively. Contrary to purely qualitative scholars, I am hoping (and optimistic) that computer scientists will one day put us out of our misery of hand coding. Thus, ultimately, I share what the authors write about CAQDAS approaches: that ‘‘the researcher always has ready access to the underlying qualitative data (e.g.,

Strikes: Sociological Aspects

The rise of industrial capitalism throughout the nineteenth century went hand in hand with the rise of strikes and other forms of labor protests. By the end of the century most industrialized countries were collecting systematic strike data. No doubt the relevance of the issue—at the heart of capitalist social relations—and the ready availability of data led to a burgeoning academic literature. What do we know? Basically, that the temporal pattern of strikes follows the business cycle: the higher the level of unemployment, the lower the number of strikes. More generally, one thing is clear: strikes are cyclical with long up-and-down swings (with the down swings typically leading to predictions of the ‘withering away of strikes’). Strikes are also linked to organization: the better organized workers are (e.g., through trade unions), the more likely they are to call on strike a larger number of workers and to be successful in their demands. Finally, strikes are linked to the political position of labor: strikes have gone down wherever and whenever labor-oriented parties acquired a stable and durable control over the government within Western democracies. But what about the future? Has the nature of strikes fundamentally changed with the shift from manufacturing to services? The evidence does show some distinctive features of service-sector strikes. Whether strikes are a thing of the past, the future will tell.

A third road to the past? Historical scholarship in the age of big data

ABSTRACT Is a third passage to the past possible, beyond Eltons and Fogels two roads of narrative history and scientific/quantitative history? One that would combine narrative historys focus on the event, on individuals and their actions, at a particular time and place, to scientific/quantitative historys emphasis on explicit behavioral models based on social-science theories? That is the question this article addresses. It illustrates a computer-assisted methodology for the study of narrative—quantitative narrative analysis (QNA)—that does just that. Based on the 5 Ws + H of narrative—Who, What, When, Where, Why, and How—QNA quantifies events without losing the event itself, without losing people behind numbers, diachronic time behind synchronic statistical coefficients. When used in conjunction with dynamic and interactive data visualization tools (and new natural language processing tools), QNA may provide a third unforeseen road to the past.

From method and measurement to narrative and number

Back in 1983, when econometrics was all the rage, freely flowing from economics even into sociology, Edward E. Leamer, a leading econometrician, wrote: ‘There are two things you are better off not watching in the making: sausages and econometric estimates’ (1983, p. 36). I was one of those sociologists, during my doctoral years at Johns Hopkins, who had been enchanted by the siren’s song of this new alchemic science: Calculus I, II, Differential Equations, Linear Algebra, Bayesian Statistics, Econometrics, Spectral Analysis (not that I did well in these courses ... but I did them, along with a great deal of computer programming in a number of languages). I ended up applying this technical apparatus to a dissertation on Italian strikes, using official statistics (strikes, but also wages, unemployment, and unionization). And where official statistics could not reach, in my drive to measure and quantify, I spent months collecting data on the number of workers subject to contract renewal in any given quarter. One of the strongest findings of my dissertation, later to become the book The Puzzle of Strikes: Class and State Strategies in Postwar Italy (CUP, 1995), is the result of that hard work of measurement: the cyclical nature of Italian strikes, the temporal ups and downs of strikes in coincidence with the renewal of labor contracts. But, ultimately, those measurements and that technical apparatus left me disappointed. The actors had all but disappeared behind statistical coefficients and ‘variables’. Where were the workers, the unions, the employers, the state? What did they do? What were their strategies, their relationships? To answer those questions, in 1981, during a post-doctoral year at the Center for Research on Social Organization (CRSO), directed by Charles Tilly at the University of Michigan, I embarked upon a new, and different project of measurement, one that would ultimately lead me closer to Aaron Cicourel’s work. Several scholars at CRSO, from Tilly to Gamson, Paige, Thomas and many graduate students, were all using newspapers as sources of sociohistorical data on actors and actions. And they were using content analysis to extract information from newspapers on those actors and actions. With my characteristic enthusiasm and lack of realism, given my computer science and statistical background, I thought I could find a much better approach to the kind of content analysis these scholars were using, one that would give me the Who, What, When, Where and Why and How not as lists of unrelated entities, as in traditional content analysis, but one that would keep the relationships between these entities fundamentally keeping intact the narrative structure of the data. And of course, in light of my quantitative background, my goal was to quantify, to turn words into numbers. It would turn out

Response: Eco, Shakespeare, Galileo, and the Problem of Textual Analysis

Umberto Eco, in a highly cited book on the role of the reader and on the semiotic interpretation of texts (1979), put it this way: ‘‘one could say that there are more things in a text than are dreamt of in our text theories. But there are also fewer things than are dreamt of’’ (Eco 1979:38). Among these fewer things Eco finds in a text is Greimas’s semiotic actantial model, the model of agents (or actants) and their roles, the basic story grammar we discuss in our paper (Franzosi, De Fazio, Vicari, this volume, 2012:1–42). But for as large a class of texts that narrative is, it is just one class, one text genre out of many. A story grammar hardly fits non-narrative texts. Yet, even these texts may contain invariant structures: Roberts’s four functional forms—description of a process, description of a state of affairs, judgment of a process, judgment of a state of affairs— applied to measure the intentions of a clause or Vicari’s modal verbs (can/could, may/might/must, shall/should/ought to, will/would) applied to analyze social movements’ different framing tasks as expressed in their texts (can/will = prognosis; must/should = diagnosis) (Vicari 2010). Depending upon what we want to see, we focus on different things in a text. Modality would not get us any closer to understanding the social relations of lynching in terms of who does what to/with whom. No more than a story grammar would allow us to understand framing tasks or intentionality. It is no doubt this wide variety of things found in texts that leaves me skeptical of the claim that codebooks and coding rules solve the problem of meaning and interpretation. If a reader brings to the interpretive act Eco’s ‘‘encyclopedia of knowledge’’ where ‘‘every text refers back to previous texts’’ (Eco 1979:19), what would Professor Popping’s codebook look like? Perhaps, the original text that Eco paraphrased in the opening lines of this reply may help us shed further light on this fraught relation between texts and methods (a new word brings up a new world). ‘‘There are more things in heaven and earth’’ says Hamlet to Horatio ‘‘than are dreamt of in your philosophy’’ (Shakespeare Hamlet

Regimes and RepertoiresRegimes and Repertoires, by TillyCharles. Chicago, IL: The University of Chicago Press, 2006. 240 pp.

The next ten chapters—the bulk of the book—present case studies (the Sharpeville massacre, the beating of Rodney King, the Abu Ghraib scandal, and so on). Each summarizes the events in question, then examines how the five methods were used in efforts to inhibit outrage. That is, these chapters illustrate how the book’s framework can be applied to a range of cases; while much of the substantive material in these chapters may be new to students, it will be familiar to older readers. The two final chapters do try to extend the framework. Martin argues, for instance, that backfire can only occur if people become outraged, and if knowledge of what has happened spreads. He also notes that there are countermoves that can circumvent each of the five methods, e.g., cover-ups can be overcome by exposing what has happened. However, most of these chapters link his framework to the literatures in social movements, social problems, and so on, so the book’s initial promise of thorough exploration is not fulfilled. Justice Ignited largely begs the most important question we might ask about backfire: under which circumstances does it occur? It is easy to point to enduring systems of repression—Jim Crow segregation, the patriarchy, and so on. For the most part, their day-to-day operations fail to elicit backfire reactions, and it is often possible to point to egregious acts of repression that do not provoke the sorts of intense, sustained outrage that characterize the cases discussed in this book. While Martin would probably respond that backfire fails to occur because the five methods of inhibiting outrage are at work, his focus on celebrated cases causes him to emphasize damage control; for instance, how did the LAPD respond once millions had seen the Rodney King video? But understanding the absence of backfire forces us to think more broadly about, say, the nature of legitimacy, the conditions that support social movements, or the media’s criteria of assessing newsworthiness. Time also seems relevant. Apartheid endured for decades after Sharpeville; should we consider backfire to be the immediate reaction to the massacre, or can we somehow credit backfire with the policy’s eventual demise? This means that we cannot simply take outrage for granted. Martin generally avoids confronting the nature of outrage. To be considered a case of backfire, which people—or how many people—must react with outrage? If a police force disciplines an officer for brutality, is it a backfire? What if reported brutality leads local activists to hold a press conference—or does it matter how much media attention results? Members of social movements often feel outrage at news reports that most people ignore—does backfire require attracting the attention of a broader audience? Does it matter whether the outrage leads to some sort of change and, if so, how soon does the change have to occur? Talking seriously about backfire requires confronting such issues. Very broad definitions of backfire or outrage make it difficult to think systematically. It might help, for example, to consider backfire as a social construction. But whose construction—the perpetrators’, their opponents’, or the analyst’s? Is it backfire when leaders of some organization think, “Oops, we’ve blundered and need to straighten things out before they get worse”? Or is it backfire when their opponents think, “The perpetrators just did something that can make them look bad”? Or must we wait for an analyst to assess whether an action had sufficiently negative consequences. We can suspect that these three things can occur more or less independently. Arguing that they are essentially aspects of the same phenomenon confuses, rather than clarifies the issue. Martin’s book reminds us that the analysis of backfire and other unexpected consequences is a worthy topic for sociological attention, but it also exposes the serious theoretical challenges the subject poses for would-be analysts.