Structuring Communicative Action There is a great deal of excitement in the sociology of culture over the possibility of formal analysis of systems of meaning and of integrating this analysis into investigations of social structure. Existing studies have examined relatively simple meaning structures using relatively simple data with fruitful results. But significantly less work has explored the benefits of formal methods for understanding complex meaning structures such as those found in philosophical texts. How can we analyze meaning structures as complex as philosophical concepts? And how can we use these meaning structures to understand the formation of cultural communities like philosophical schools? This paper proposes an analysis of semantic relations based on copresence in semantic neighborhoods. In the following, I employ this method to model the conceptual structure of Jürgen Habermas’s Theory of Communicative Action. I then build on this method to propose a systematic approach to analyzing similarity and difference among conceptual structures created by different theorists—that is, authors’ cultural relations to one another. Examining patterns of cultural relations alongside patterns of social relations, we can analyze how cultural communities such as philosophical schools develop. I formalize texts by Max Horkheimer & Theodor Adorno, Jürgen Habermas, and Axel Honneth—the three generations of Frankfurt School Critical Theory. The convergence of their meaning structures reveals how the Frankfurt School’s “third generation” established the school by merging previous generations’ disparate ideas in its own. Thus, combining natural language processing and network analytic techniques, I introduce a powerful, yet flexible, tool for investigating structures of shared culture. Keywords: Computational Text Analysis, Formal Text Analysis, Culture & Networks, Sociology of Philosophy, Intellectual Schools, Frankfurt School/Critical Theory 1. Introduction Formal analysis has been one of the most fruitful approaches to studying the symbols, meanings, and identities that comprise culture. Because meaning inheres in language, much of this approach has focused on textual analysis. Pioneers in the field such as John Mohr have used Galois lattices and blockmodeling to analyze texts’ latent conceptual structures, including the duality between social identity categories and the provision of social welfare services (Mohr 1994; Mohr and Duquenne 1997) and the relational logic linking social identities and diversity practices (Mohr and Lee 2000, Mohr, Bourgeois, and Duquenne 2004). Researchers like Kathleen Carley have focused on the semantic properties of texts, exploiting the meaning that 1 inheres in grammatical constructions and thus preserving in formal models texts’ richness (e.g. Carley 1993). Despite their significant successes, these formal techniques may be improved in two respects. First, they are useful primarily for examining texts with relatively simple meaning structures, unable to capture meanings in complex texts. Second, extant formal methods are amenable to analyzing relatively small text corpuses; they are not designed for use on larger corpuses. Thus, there remains a challenge, on the one hand, to delve into more complex meaning structures, and on the other hand, to scale up these studies. The study of cultural institutions—social organizations founded on shared cultural ideas and practices—could benefit greatly from confronting these challenges. Research in this area attempts to integrate the study of cultural phenomena—often using formal models—into the analysis of social structure (e.g. Mische, 2007). But a simplistic treatment of culture has been an obstacle to understanding these cultural-structural entities. Current research does not integrate into empirical analysis much of what we know about the complexity of cultural meaning. This prevents us from studying the strength and quality of “cultural relations” within institutions as well as important institutional processes like the evolution and negotiation of meanings. How can we better understand the complexity of culture and the cultural processes fundamental to institutions? Following John Mohr, I argue that the answer is— counterintuitively—to code less. When analyzing cultural data, we should avoid arbitrary coding schemes and find a way to “get the entire text into a computer” (Mohr 1998: 366). That is, existing formal methods operate by sorting text objects into a predetermined structure of cultural categories (e.g. Franzosi’s (2004) pathbreaking work with subject-verb-object triplets or Mohr & Lee’s (2000) analysis of identities and institutional action). While our instinct may be that more or different categories are the key to embracing the complexity of culture, this view is 2 problematic. No matter how imaginative the categories, coding still limits the meaning of a text to the analyst’s preconceived assumptions. To better embrace complexity, we should formalize meaning structures without immediately confining them to an ontological framework. The analysis of semantic relations based on semantic neighborhoods performed here takes up this task. It creates reliable meaning structures out of complex texts without coding and it does so on a large scale. These structures then form the foundation for an analysis of institutional processes. In the following, I propose semantic relational analysis for the study of cultural meaning and of socio-cultural processes in four steps. First, it will be necessary to define semantic meaning—a type of meaning that is useful for and accessible to formal analysis. Second, the idea of semantic relations based on co-presence in semantic neighborhoods will be introduced. I perform over 300 robustness tests to determine the appropriate criteria for identifying semantic relations—in other words, the best semantic neighborhood for analysis. The co-presence of concepts in paragraphs appears to produce the most robust meaning structures. Third, the paragraph co-presence method will be used to retrieve the semantic meaning of Habermas’s Theory of Communicative Action. The meaning structure produced successfully captures the text’s main argument and main interlocutors: Max Weber’s concept of instrumental rationality, which pervades the contemporary (Parsonian) social system, has supplanted the lifeworld’s logic of communicative action. The challenge for modern society is to break free from the system through rational communication between human beings. Other general features of semantic structures will be examined to uncover key argumentative moments in a text. Finally, I will demonstrate how semantic meaning structures can be used to analyze the development of cultural communities such as philosophical schools. Beyond examining individual texts by themselves, semantic relational analysis’s most distinguishing feature is that it creates data that 3 allows researchers to systematically analyze similarities and differences among multiple texts. Texts formalized as networks can be intersected to reveal these cultural relations among their authors. And examining these patterns of cultural relations alongside social relations can reveal how cultural communities are formed. Analyzing the relations among multiple meaning structures, I will reveal how Critical Theory was established as a philosophical school. It has long been argued that the “three generations” of Frankfurt School Critical Theory were united by a common interest in the human capacity for “reason” (Anderson 2000). A semantic relational analysis of their three “reason” structures confirms intellectual historians’ hypothesis that the third generation of the Frankfurt School (Axel Honneth) was responsible for forging the philosophical unity underlying this tradition. He did this, I demonstrate, by bridging a “cultural hole” (Pachuki & Breiger 2010) between his predecessors. 2. Semantic Meaning The formal analysis of text does not imitate subjective reading and interpretation; rather, it is a wholly different approach to textual analysis that accesses a distinct type of meaning. Thus, as a first step in demonstrating semantic relational analysis, it important to define provisionally the type of meaning that it uncovers. I propose that it uncovers the meaning that inheres in words’ network of relations to other words and I will refer to this type of meaning as semantic meaning.1 Semantic meaning relies chiefly on structuralist theories of meaning, such as Saussure’s 1916 [1974] concept of syntagmatic meaning and Charles Morris’s (1938) syntactical Saussure and Morris’s approaches are earnestly “syntactic,” focusing on how word order and grammatical rules link words to construct the meaning of larger symbols like sentences. Because I am not similarly interested in syntax but in words’ relationships to their semantic contexts more broadly, I will refer to the meaning rooted in the relationship between signs by a more comprehensive term: semantic meaning. This is, however, still distinct from Charles Morris’s (1938) own definition of “semantic meaning” as the relation of objects to signs. Rather, I deploy the term “semantic” in its more general usage as many computer scientists do—as the study of how meaning arises from the relations between linguistic symbols. 1 4 meaning, that contend that a word cannot be understood apart from how it is used in conjunction with other words. They imply that meaning does not reside in a single word but in the network of relations between words. German Hermeneuticians expressed a similar idea as they proposed an alternative way to explain accounts of historical causality. Following Schlegel (1808), Dilthey’s (1883) concept of a “meaning nexus” (Sinnzusammenhang) is a constellation of nonindependence bound by meaningfulness. Humans understand historical objects and events by situating them in broader contexts of meaning, assessing them in their relation to the other objects and events that occur contemporaneously with them. These formulations suggest that we may be able to reproduce such a meaningful constellation by graphing word co-occurrences. The meaning of a given concept may be discovered by examining its network of relations with other words. Focusing on semantic meaning enables us to broaden the methodological possibilities for cultural research, for semantic meaning can be retrieved from texts with computational methodologies that have remained foreign to the analysis of complex meanings.. Computational methods can identify and analyze the meaning relations upon which possible interpretations are based, and generate from them structures of semantic meaning. We may analyze how meanings converge in institutions like philosophical schools by uniting diverse meaning structures into larger constellations that highlight their relation to one another. Imagine two meaning structures, S and T, that represent the ideas of two different individuals. S ∪ T is the constellation of meaning structures. Their intersection, S T, represents how the meanings are similar, and the elements left out of that intersection represent how they diverge. Examining either S or T alone, we see the confluence of meaning from a second perspective. Each singular expression of meaning (S or T) is itself a constellation—a 5 unique intersection of diverse influences and ideas brought together by the thinker in question. Each shared fragment (S T) is an embolum (Peirce [1940] 1955), a sign thrown into another. A portion of S is in T; a portion of T is in S. These embola have a sociological significance: they represent cultural associations and divisions among individuals, forming the basis for what much recent research refers to as “cultural ties” or “cultural relations” (e.g. Mische 2007; Pachuki & Breiger 2010). S may be an embolum in T because T is engaging with S to either elaborate on or oppose S. With the ability to investigate cultural relations we can investigate the patterns of intellectual engagement underlying the formation of philosophical schools. For example, it is possible to analyze how ideas are passed down from teachers to students or contested by contemporaries. So while semantic meaning is not the type of meaning that philosophers and linguistic pragmatists seek, I argue that it is a solid foundation for a sociological investigation of meaning. It allows us to analyze cultural meaning as a distinct type of relation between individuals and thus helps us uncover the social processes in which meaning is produced. 3. Testing Semantic Neighborhoods Meaning relations are not computational text analysis’s usual object of study. Rather, the classic computational approach is traditional content analysis (TCA), which tracks the frequency of concepts within and between texts. TCA was introduced originally by Stone et al.’s The General Inquirer (1966). It focuses primarily on counting the existence or frequency of a word or words in a text corpus over time or between authors, often claiming that the words of interest signify a larger cultural category (e.g. Fan 1988; Namenwirth and Weber 1987; Garson 1985; Sullivan 1973; Moretti 2005, 2009; Evans 2010). Recently, TCA has been brought to the forefront of cultural research by the “big data” text mining movement, which seizes on the expansive data offered by ongoing book digitization projects. Jean-Baptiste et al., for instance, 6 studied the frequency of n-grams2 in 5.2 million books over the course of over 500 years to expose linguistic changes and shifts in collective memory (2010). Lieberman et al. studied the regularization of English verbs over 1,200 years in a 17.9 million word corpus (Lieberman et. al 2007). The benefits of TCA are substantial. First, its quantitative semantic approach generates insights into meaning unconstrained by the subjective interpretation of only one person. Second, because it is computer-assisted, it allows for research that embraces massive corpuses. But developed in Computer Science rather than in the humanistic disciplines, TCA often lacks sensitivity to the complexities of language. Its most significant weakness is its assumption that words have a single stable definitional meaning. When a word is counted, TCA treats the word as meaningful in itself, and by that logic, as having the same meaning every time it appears. But these assumptions are incompatible with what we know about language. The same word can mean various different things depending on its context. For instance, when we count the frequency of the ambiguous term “culture” in a corpus of texts, we may be counting mentions of the opera and theater, the laziness and anti-intellectuality of Americans, or the stuff growing in a petri dish—some of what we want, some of what we do not. Counting “culture” is a flawed method because “culture’s” meaning is neither stable nor contained in the word itself.3 Rather, its meaning arises from the relationship between words, its meaning-rich semantic context of which TCA takes no account. This is the insight captured by formal analysts of text, who shift textual analysis from an ontology of concepts to an ontology of relations. Texts are not simply “bags of words;” words occur in a certain order; they are embedded in other words, and these semantic constellations are 2 An n-gram is a sequence of words with length n. It is worth mentioning that some improvements have been made to enhance TCA’s reliability. For example, Krippendorff (2004) has developed strategies to improve intercoder reliability, and Franzosi (2004) has worked to to increase semantic coherence. 3 7 more meaningful than the words in and of themselves. Classical structuralist theorists have long taken this position. French linguist Ferdinand de Saussure provided perhaps the earliest articulation. He argued (Saussure [1916] 1983) that meaning arises from the systematic distinctions that differentiate words (as signs or sounds) from one another. The interpretation of meaning is connected to the system of relations that link cultural objects. His thought was further developed by Levi-Strauss (1963), who proposed deep structural principles that organize the complex system of relational patterns. From a Levi-Straussian viewpoint, a woman “means,” for instance, which families her marriage unites. Georg Simmel ([1907] 2004) shared this perspective when he argued that the situational juxtaposition of objects in exchange is the source of its value. $70 “means” the ability to exchange it for a sweater or for a pair of sneakers when they are declared equivalent in an exchange relation. Contemporary researchers have studied patterns of conceptual association in texts in various ways. A first approach uncovers texts’ latent conceptual structures. John Mohr has been a major figure here, using Galois Lattices and blockmodeling to discover the meaning inherent in the dual association between identity categories and poverty relief services (Mohr 1994; Mohr and Duquenne 1997). Ronald Breiger (2000) uses correspondence analysis and Galois Lattices to cross-classify symbols and uncover Supreme Court justice’s control over and interest in areas of legislation. John Levi Martin (2000) uses an entropy-based dispersion measure to match occupational categories to animals in “What do Animals do All Day?” to uncover the system of social stratification among species. Peter Bearman and Katherine Stovel (2000) model texts as nodes and flows and use network techniques such as density, distance, and centrality, to contrast the narrative structures of becoming and being a Nazi. 8 A second major approach to formal text analysis focuses on a text’s grammatical constructions. Researchers in this tradition examine words’ grammatical functions, contending that their semantic-structural equivalence indicates that they occupy a common conceptual space (e.g. Ignatow 2004, Cerulo 1998). Roberto Franzosi (2004), for instance, has written extensively on semantic triplets of subject-action-object. A particularly promising method is map analysis (Carley, 1993, 1994, 1997), which relates concepts in a text to each other through semantic relations that vary in strength, sign, direction, and meaning. Kathleen Carley, for instance, has used map analysis successfully to show how the cultural perception of robots has shifted from dangerous to lovable since the 1950’s through a map analysis of 30 science fiction novels (1994). These formal analyses, however, have been limited in two areas. First, formal analyses have only been performed on relatively simple textual data; their ability to analyze complex texts remains untested. The best known works analyze pre-sorted data (Mohr 1994: Breiger 2000) or conceptually simple narratives (Martin 2000). But many texts contain ideas and meanings that are far more nuanced and contested. Philosophical texts, for example, can rarely be analyzed in terms of objects with simple attributes. Rather, each work of philosophy explores a set of main concepts by continually embedding them in diverse conceptual contexts. Through considering many such embeddings, we may finally come to a comprehensive understanding of a concept. Second, extant formal methods lack the explanatory power of TCA because they tend to rely on painstaking, meticulous hand-coding. The fine-grained analyses that have become characteristic of this sub-field are time prohibitive for large text corpuses. However, there appears a desire to scale up these studies to embrace larger text corpora (and draw macro-level conclusions). Roberto Franzosi, for instance, makes textual data amenable to statistical analysis by applying to his semantic grammars a set theoretical framework (2004). John Mohr (1994) 9 uses automated n-gram searches to create categories of poor relief organizations and clients that can be matched and analyzed with blockmodels. But even these approaches offer only the ability to analyze qualitative data quantitatively; they do not alter or eliminate the lengthy coding process. In response to these challenges, I introduce a “big data” text mining approach that espouses formal text analysis’s attention to semantic context. Computer-assisted, it has enormous reach, allowing researchers to analyze large text corpuses. But by modeling texts formally as semantic networks, it also focuses on the relations between words, preserving some of texts’ complexity meaning. This method will be used to analyze Juergen Habermas’s Theory of Communicative Action, whose extended length and high degree of analytical clarity make it an excellent text for methodological testing. The remarkable ability of this analysis to schematically reproduce the main argument of this text confirms the potential of formal methods to capture complex meaning structures. 3.1 A semantic relation is defined as co-presence in a semantic neighborhood. The method relies on the discovery and analysis of semantic relations embedded in philosophical texts. While a semantic relation can be defined in many ways, I focus on one possibility: the co-presence of words in a semantic neighborhood. A semantic neighborhood is a more inclusive version of what Carley et. al’s (2013) AutoMap software calls a “stop unit.” It connotes a unit of text that may be demarcated in various ways. Like a stop unit, a neighborhood may be grammatically determined, such as a comma-separated clause, a sentence or a paragraph, but it may also be determined by length without regard for grammar, such as a window of 3, 10, or 15 words. Words are considered related when they are co-present in the same neighborhood. 10 This semantic method therefore follows the logic of Ronald Breiger’s (1974) famous argument that people can be thought of as related to each other when they share common group memberships. Semantic relations may in fact be the example par excellence of the duality of shared group membership and direct relation. Words that occupy the same lexical units contextualize and define each other. Every utterance is a string of words, each of which would have no meaning, had it not been given meaning by its position among others.4 But without a valid definition of “group,” this argument about words (as well as the argument about people) holds little water. One could argue, for example, that every word in a book is related to every other. This may be true in principle, but because a book can be so long, many of the semantic relations within it may be so weak that they are practically meaningless. Thus, the first task of this study is to examine the semantic relations implied by common membership in various semantic neighborhoods and find the type of neighborhood that constitutes a component of a text’s core conceptual structure. This will be the lexical unit in which semantic co-presence identifies the most semantic relations (minimize false negatives) that are all meaningful (minimize false positives).5 3.2. The paragraph is the most reliable semantic neighborhood. I compare the semantic relations captured by 301 different semantic neighborhoods into which Habermas’s Theory of Communicative Action can be divided: clauses separated by punctuation, sentences, paragraphs, and n-grams from size n=3 to n=300. The first three analyses, then, are based on units provided by the grammar of the English language. These units 4 I do not analyze or discuss specific types of semantic relation in this study. Those types may include but are not limited to synonymy, antonymy, hypernymy, hyponymy, meronymy, holonymy, and troponymy. 5 An analysis produces false negatives when all it fails to capture all meaningful semantic relations for a given concept. It produces false positives when it identifies semantic relations that are not actually valid. 11 are likely the most meaningful because each represents a single coherent expression. That is, according to the rules of the English language, words co-present in such phrases receive meaning from each other and define each other relationally. The next 298 analyses are for moving windows of text—n-grams—of size 3 to 300. These tests allow greater insight into how the mere size of the text-chunk affects the results of the analysis and helps me determine which lexical unit appears most reliable. The procedure for discovering the most appropriate semantic neighborhood is as follows. First, I digitize the books by scanning them and running optical character recognition software on them. Then, for each analysis, a computer program separates the text into the desired semantic neighborhoods. A second program then drops all the stopwords (articles, many prepositions) from the analysis6 and establishes a relation between words present in the same units. The result is a relational matrix similar to any matrix a network analyst might examine. Semantic relations are considered notable when they are “strong”—that is, when they occur frequently in a text. By acknowledging only the relations occurring most often in the text, we can filter out the “noise” that comes from offhand remarks, anecdotes, passages from other works cited in the text, and focus on its core conceptual structure. Although focusing on only strong relations might produce some false negatives, its ability to substantially reduce false positives justifies the decision. More detailed discussion of clauses, sentences, and paragraphs, as well as full results of these analyses, is presented in Appendix A. I start with clauses, sentences, and paragraphs. The three methods produce surprisingly similar results, but the variations that do occur between methods follow a discernible logic. With 6 Although stopwords such as prepositions may be important for producing meaning, removing stopwords is advisable for this particular semantic relational analysis because I am searching for texts’ philosophical as opposed to their grammatical meaning. That is, I am not interested in a deep linguistic analysis of a particular sentence or phrase; I am interested in the way in which primary philosophical concepts are nested in and contextualize one another over a longer text or set of texts. 12 each move from a smaller to a larger unit—that is, from clause to sentence and from sentence to paragraph—there are four patterns of note: 1) simple bigrams are replaced with semantic relations bridging greater distance. 2) The graph produced becomes a more coherent structure; semantic cliques previously separate from the main structure become integrated into it. 3) Semantic “hubs”—central keywords—become increasingly evident. 4) The frequency of relations increases substantially. These four patterns can be considered advantageous because they move the structure toward greater coherence, reliability, and intricacy. But with these advantages comes at least one drawback: dominant nodes become amplified, possibly obscuring other structural patterns. Despite this drawback, the test of grammatical units suggests that the paragraph method appears to produce the most useful data for further analyses. Because the paragraph is the grammatical unit that represents most accurately a complete and coherent thought, the paragraph level yields semantic structures that are composed of reliably strong semantic relationships, minimizing false positives. At the same time, its reach over a larger span of text minimizes false negatives, producing semantic structures that are dense and complex enough to facilitate intricate analyses. The second set of analyses was the “n-gram test.” The purpose of the n-gram analysis is to get an impression of how the mere size of the lexical unit affects the meaning structures formed. An n-gram is simply a moving window of text without consideration for grammatical units. If a 3-gram is my grammatical unit, any two words used in the same 3-word chunk (i.e. words next to each other and within two words of each other) will be considered to have a semantic relation. I examine the relations captured when the n-gram is of size 3-300, noting the thirty that occur most frequently for each size. 13 Because small n-grams tend to err on the side of false negatives while large ones tend to err on the side of false positives, semantic relations that are textually closer are weighted more heavily than those that bridge greater textual distance. Thus, more distant relations that are potentially false positives must be extremely frequent to confirm their meaningfulness and enter the results of the analysis. I start with the smallest size (3-gram) and examine the structure composed by the semantic relations it implies; as I increase n-gram size, we should expect new semantic relations that bridge greater textual distance to appear in the top-thirty semantic structure, replacing ones that are textually closer but substantially less frequent. I seek the point at which using a larger n-gram no longer introduces relations that bridge greater textual distance but are so frequent that they must be reliable. This is, in other words, the point of equilibrium where a tendency toward false negatives is replaced by a tendency toward false positives—where false negatives and positives are both at their lowest levels. More detailed explanation of weighting and full results of the N-gram tests (Figure. A.4) are presented in Appendix A. I found that n = 104-146 is this critical moment in the analysis of Theory of Communicative Action. This key n-gram is closest in size and results to a paragraph (the average paragraph length is 108 words). This n-gram is also the most similar to a paragraph in terms of its results, with only two low-ranking semantic relations differentiating the two structures. Thus, we can conclude that that the paragraph size of analysis has relatively few false negatives and false positives—it locates as many semantic relations as possible that are all meaningful and thus, offers the greatest possibility for revealing the core conceptual structure of the text. The following analyses will therefore employ the paragraph method. 7 7 It is important to note that Theory of Communicative Action, originally written in German, is analyzed here in English translation. Discussion of the advantages and disadvantages of analyzing these texts in translation can be found in Appendix B. 14 4. The Structure of Communicative Action Semantic relational analysis’s ability to reproduce schematically the core argumentative structure of Habermas’s text is highly encouraging. Briefly, in the Theory of Communicative Action Habermas employs a thorough critique of the individualism inherent in Weber’s theory of purposive rationality and Parsons’s notion of the social system. Purposive-rational action pervades the social system, which has supplanted the lifeworld, the domain of rational communication among human beings. The challenge for modern society, then, is to break free from the modern social system by means of rational communication. Figure 1 is a graphic representation of the Theory of Communicative Action. Semantic relations displayed in the graph occur at least 400 times in the text. The most central concepts are exactly that—the main topics addressed the text. The most important is clearly action, followed by social, theory, and system. I will proceed by examining the graph intuitively and reading it against my own understanding of Habermas’s text to show how it captures the text’s main philosophical ideas. Figure 1. The Structure of Communicative Action8 ---------------------Figure 1 about here ---------------------Notice first of all Habermas’s main interlocutors in the graph—Weber and Parsons. In the upper right hand quadrant of the graph, Weber’s theory of rationality is intimately linked to Parsons’s theory of the system. That is, Weber’s rationality pervades the type of social system described by Parsons. To this cluster’s immediate left, the concepts of the social world and 8 In all graphs, nodes are positioned manually to make edges and nodes more clearly visible. In this first graph of Theory of Communicative Action, the strength threshold for edges has been dropped to >400 co-occurrences. 15 society bridge discussions of Weber, Parsons, and the system, with the communicative action in the lifeworld. That is, our social world or society can be conceived of simultaneously as a system and as a lifeworld. We can tell which side Habermas takes in this opposition; he critiques Weber and Parsons in favor of communicative action in the lifeworld. While the discussion of Weber and Parsons occupies a small and isolated portion of the structure, the majority of the graph falls in favor of communicative action in the social lifeworld, particularly the ideas of speech acts and validity claims as forms of action. The position and centrality of the concept system and the majority of the concepts in the structure therefore elucidate the main problem of society as Habermas sees it: we are trapped in this Weberian-Parsonian system, and we must break free from it through communication between human beings—the activity of the lifeworld.9 We can go beyond this general overview and examine the book’s argument more deeply by disaggregating it into its constituent chapters. Semantic structural features reveal the argument’s construction. I examine the exploration and integration of main concepts as well as a text’s style of interlocution—how the author engages with the ideas of others. I focus on a few chapters, which I chose because they illustrate these semantic structural features particularly well. 4.1 Lower density between points of high betweenness indicates a weaker integration of ideas The first and simplest thing these structures expose is the exploration and integration of main concepts. Rhetorical features are represented by structural “hubs” and the connections 9 I do not claim that one can achieve this theoretical interpretation by merely looking at the semantic structure. This fluid interpretation, while conforming well to the structure, comes from my own knowledge of the text. The automated analysis cannot produce coherent articulations because the existence of semantic relations in the graph provides no clues as to the type of semantic relation it represents. That is, there is no way to tell whether concept A is concept B, concept A opposes concept B, concept A is a type of concept B, or concept A modifies concept B, etc. 16 between hubs. When words are in a position of high betweenness10, it means that they are central concepts being explored in the philosophical text. A text can be understood as the nuanced linking of these central concepts. Density11 of linkages between hubs reveals the strength of the text’s conceptual integration. When concepts are densely integrated in a text, it means that the argument is coherent and well developed; the author successfully makes a new idea out of existing ones.12 When concepts are loosely integrated, it means that the argument is less developed. It is fragmented into sub-arguments, where a new idea is hardly produced out of existing fragments. The first chapter, “Introduction: Approaches to the Problem of Rationality,” exemplifies weak conceptual integration. The central concepts are action and world, which are linked through topics of communicative rationality, understanding, and validity claims. Habermas strives to merge a theory of rational action with man’s understanding of the world. In the history of philosophy, he argues, reason and a certain conception of the world have been analytically inseparable. Man has always employed his reason to find a conception that unifies the world’s multiple appearances. Habermas wishes in this text, however, to reject notions of transcendental reason and holistic worldviews. He proposes instead to destabilize these traditional concepts and to redraw the linkages between historically evolving rationality structures and the various ways of understanding the world that provide rationality’s conditions of possibility. Figure 2. Chapter 1: Introduction: Approaches to the Problem of Rationality ---------------------- 10 Betweenness centrality of a given vertex is the number of shortest paths from all vertices to all others that pass through that vertex. In short, it is the tendency for that vertex to connect or serve as a go-between between two others. 11 Density is the proportion of all possible dyadic ties that are actually present, how “well-connected” a network is. 12 When maximized, density between two hubs leads to those hubs’ erasure; they merge into one dense cluster representing a new ideational synthesis. 17 Figure 2 about here ---------------------As an introduction that merely raises the question of how one might merge these ideas, actual conceptual integration is remarkably weak. The “spokes of a wheel” structures are obvious, indicating that both action and world are main concepts, but their relation to each other is thinly elaborated. We know the integration will have something to do with rationality as communicative action that emerges as validity claims, establishing understanding between individuals, but a coherent theory of the linkages between rational action and various understandings of the world has not yet been formulated. 4.2 Higher density between points of high betweenness indicates a stronger integration of ideas By contrast, main concepts are better integrated into a coherent argument in chapter six, “Intermediate Reflections: System and Lifeworld.” Here, Habermas introduces and integrates two of his primary concepts—competing conceptualizations of society as a lifeworld and as a social system. The lifeworld is the realm of lived experience and social practice, characterized by communicative action and mutual understanding among interacting individuals. Knowledge is the binding force of a lifeworld; it is the shared stock of background convictions that enable shared understanding among its members. At the same time society can be conceived of as a social system—an economy composed of market forces and a state administration governed by instrumental rationality. Interaction as communication and shared meaning do not take place here; rather, this social system undergoes processes of system integration and system differentiation. Figure 3. Chapter 6: Intermediate Reflections: System and Lifeworld 18 ---------------------Figure 3 about here ---------------------Habermas elaborates these two fundamental ways of conceptualizing the social order in themselves, but more importantly, their interdependence. Strong integration of these two conceptions shows in the perfect social-system-society-lifeworld-action pentagram clique. That is, a potential “two-wheel” pattern has been integrated to the point where it melds into a single cluster of perfect density. So although Habermas discusses at length the system and lifeworld’s historical decoupling, his analytical emphasis falls on their important interchange—the way the system’s instrumental rationality imperatives “make their way into the lifeworld from the outside—like colonial masters coming into a tribal society—and force a process of assimilation upon it” (Habermas 1981: 355). 4.3 Betweenness of interlocutors in meaning structures exposes styles of interlocution A second rhetorical feature that semantic relational analysis can highlight is the styles of interlocution in a text. This is the way an author engages with the work of others. As a first attempt, we can decipher whether interlocution is shallow or deep—the extent to which interlocutors are well-discussed and well-integrated into the text. We detect this in the betweenness of an interlocutor’s node, as well as the density of connections between the interlocutor’s cluster and the rest of the semantic relational structure. Chapter five, “The Paradigm Shift in Mead and Durkheim: From Purposive Activity to Communicative Action,” exemplifies instances of both deep and shallow interlocution. Here, Habermas engages with George Herbert Mead and Emile Durkheim to explain how communicative social action is at 19 once symbolically mediated and normatively regulated. The depth of his engagements with the two differs remarkably. Habermas’s engagement with Mead is deep. In fact, Mead begins the synthesis that Habermas pursues; symbolically mediated interaction through an exchange of gestures between a first and second organism must adhere to rules that guide behavior and interpretation of meaning. But Mead’s ideas alone do not complete Habermas’s philosophical synthesis; Durkheim provides an important piece. Only through appeal to Durkheim’s theory of the sacred and moral norms can we understand how systems of linguistic meaning delineate communities—how linguistic norms can serve as the morality that binds and distinguishes between social worlds. Habermas’s engagement with Durkheim, however, is perfunctory and shallow. This is evidenced by the “chain” of Durkheim stretching away from the rest of the graph. To be sure, the reference to Durkheim allows him to think about the moral as the sacred, but its elaboration is set apart from the other fundamental components of Habermas’s theory; it has little use otherwise. Figure 4. Chapter 5: The Paradigm Shift in Mead and Durkheim: From Purposive Activity to Communicative Action ---------------------Figure 4 about here ---------------------Habermas is borrowing and cherry-picking from Durkheim, not truly engaging. By contrast, the engagement with Mead is well embedded into the larger meaning structure. Mead is the high betweenness node that allows Habermas to integrate symbolically mediated interaction with communication, attitudes, action, behavior, communicative, theory and meaning. Semantic relational analysis shows that this is a much more fruitful engagement, one that is undeniably crucial to the core of Habermas’s theory. 20 5. Structuring the “Reason” Problematic: Cultural Ties and the Formation of Schools In the analysis above, we have captured and analyzed a text’s semantic meaning. The ultimate goal of this study, however, is to use this textual data to contribute to sociological knowledge. The next task is to consider how semantic meaning is part and parcel to social relations and social organization, and relatedly, how semantic structures help us investigate socio-cultural processes like the establishment of cultural institutions such as philosophical schools. This starts by analyzing multiple semantic meaning structures. Such analysis produces data on “cultural relations” that can be integrated easily with social structural data. When we analyze the relation between sets of cultural and sets of social relations, we can effectively theorize cultural processes. 5.1 Relating culture to social structure The way in which cultural meaning relates to social structure is a core concern for the Sociology of Culture. Although classic studies in network analysis treat culture and structure as entirely distinct (Katz and Lazarsfeld 1955; Coleman, Katz, and Mendel 1957; Burt 1987; Kim and Bearman 1997), the emerging consensus is that social structure and culture are “dual”— inseparable and co-constitutive. That is, individuals’ social relationships to one another indicate as well as foster shared beliefs, tastes, and identities. A first approach demonstrates how social structure fosters shared cultural beliefs. Ann Mische’s work exemplifies this perspective, describing networks as bound by loyalties, solidarities and identities (2007). Common ideas, identities, and beliefs are forged within networks as understandings emerge from interactions among people. Dense, strong-tie networks breed intense commitments and ideals (Della Porta 21 1988). And when these networks overlap, ideals and beliefs can be transferred from one domain to another and strengthened in the process (Mische 2003, 2007, Baldassari & Diani 2007). At the same time, however, strong divergent social pressures on individuals can impede the creation of broad cultural consensus (McAdam and Paulson 1993). Recent scholarship has broadened the duality argument, showing how common tastes and beliefs influence social organization; culture can also forge social relationships. Omar Lizardo (2006), for example, argues that shared “high brow” cultural tastes are more easily “converted” into strong tie networks, while tastes for popular culture tend to form weak ties. Stephen Vaisey and Omar Lizardo (2009) then extend this logic into the realm of moral values, arguing that shared moral worldviews are the basis of friendship relations and the effort (or lack thereof) to sustain those friendships over time. In distinguishing between and comparing the interpersonal and the cultural dimensions of social relationships, such research suggests that it may be analytically useful to treat shared culture as a type of relationship in itself. Mark Pachucki and Ronald Breiger (2010) take this step as they discuss the cultural ties that bridge “cultural holes,” as do Schultz and Breiger as they explore “The Strength of Weak Culture” (2010). An analysis of multiple semantic meaning structures builds on this line of inquiry by emphasizing the intricacies of cultural phenomena. It is essentially an investigation of cultural ties—like many approaches, a relationship between cultural objects is used to index a cultural relationship between their creators. But by analyzing cultural objects in their complexity, we can examine cultural ties in their complexity. The multiple fine-grained cultural convergences that comprise a cultural tie between individuals represent how and how much individuals are in cultural agreement (in much the same way as Carley & Palmquist’s (1992) comparison of cognitive maps). This is valuable for sociological analysis because finer-grained changes in 22 degree and type of cultural relations can better illustrate cultural processes like influence, imitation, or the establishment of institutions, than can coarse binary measures of cultural relatedness. 5.2 “Reason” through three generations of Critical Theory An analysis of semantic relations helps us examine the cultural ties among Max Horkheimer/Theodor Adorno, Juergen Habermas, and Axel Honneth, leaders of the so-called three generations of Frankfurt School Critical Theory. I ask whether the conceptualization of the school as unfolding over three generations is justified by the actual patterns of intellectual continuity these authors exhibit. Historians of the Frankfurt School have long debated whether the school should be understood as developing in “generations,” and if so, how many generations there are. Many scholars accept the thesis that there are three generations. Anderson (2000), for instance, argues that there are three generations as he situates Axel Honneth’s work among his predecessors.’ And Honneth (2007), the purported leader of the third generation, has been explicit about his attempt to continue the school’s tradition. But other scholars have opposed this characterization. Some suggest that the school ended with Habermas; only the first and second generations are truly part of the school. Postone (2009), for example, acknowledges only the first generation and Jürgen Habermas as he analyzes the historical trajectory of Critical Theory. Still others emphasize Habermas’s break with the first generation, implying that there are no generations at all; the school began and ended with Horkheimer and Adorno’s circle. Habermas (1985: 209) himself denied that he ever attempted to continue the tradition of the school. Likewise, some of Habermas’s critics such as Gerhard Bolte’s circle (1989) and Heinz Steinert (2007) emphasize the theoretical discontinuities between the first generation and Habermas. 23 Here, I test the hypothesis that there is theoretical continuity between these supposed three generations of the Frankfurt School. I focus my analysis on the three generations’ similarity in their respective theories on a single concept—the human capacity for reason. It has often been claimed that the tradition of Critical Theory is united by a concern for the concept of reason (e.g. Honneth 2007). Critical theorists argue that societal critique and social liberation depend on the ability to employ reason in a society entrenched in the irrationality of the capitalist economy and the modern state. Thus, the great task of Critical Theory is to develop greater understanding of the conditions under which the human capacity for reason can be exercised. If there is theoretical continuity among these three generations, I propose, it would be uncovered by an analysis of each generation’s development of the concept of “reason.” Through intersecting semantic meaning structures, I show how the “reason” problematic united three generations of Critical Theory. The analysis shows that a unified theory of “reason” and thus the school was forged through the third generation’s bridging of cultural holes.13 Because I am interested in these authors’ converging understandings of a specific concept (“reason”) rather than the three books in general, I do not analyze here the general conceptual structure of each text, as I did previously with Habermas’s Theory of Communicative Action. Instead, I form ego-centered semantic networks around the concept of “reason.” That is, meaning structures will feature “reason” and the 16 concepts most often associated with “reason” over the course of each text. For this last analysis, I apply semantic relational analysis to Horkheimer/Adorno’s Dialectic of Enlightenment (1947), Jürgen Habermas’s Theory of 13 Pachucki & Breiger (2010) have also described the process of “bridging cultural holes.” They argue that because structure and culture are “dual” facets of social networks, bridging structural holes as in Burt (1992) is culturally contingent. That is, one may often bridge a structural hole through a cultural tie. I define the term “cultural holes” differently. Having analyzed the Frankfurt School as a distinct set of social and distinct set of cultural ties, a “cultural hole” is a discontinuity in the structure of cultural relationships. These cultural holes are especially interesting, I argue, when there is a cultural hole where a social relationship exists—an instance in which two persons have an interpersonal relationship with one another, but differ substantially in beliefs and interests. 24 Communicative Action (1981), and Axel Honneth’s Critique of Power (1991). I locate the commonalities in these meaning structures by intersecting semantic networks. A graph of the constellation of meaning structures notes their convergences. This cultural data will then be considered in light of the authors’ social relationships to one other: they form a perfect teacherstudent-chain where Habermas was Adorno’s assistant and Honneth was Habermas’s assistant. A lineage in Eviatar Zerubavel’s (2012) sense, there are direct ties between successive generations, as well as an understood ancestral tie between Horkheimer/Adorno and Honneth, despite the fact that Horkheimer and Adorno died before Honneth had the chance to learn from them directly. 5.3 Honneth’s third generation of Critical Theory bridges a hole and establishes the “school.” The problem of “reason” is redefined by each generation of Critical Theory. Briefly, the “reason” problematic was initially understood by Max Horkheimer/Theodor Adorno as a project for the individual consciousness. It was then recasted as inhering in communication by Habermas, their student. And in the end Habermas’s junior colleague, Axel Honneth, unites these opposing strains of thought by offering a third, hybrid theory of “reason”—as a process with both individual-reflective and communicative facets. Results from these analyses are summarized as adjacency matrices in Figures C.1, C.2, C.3, and C.4 in Appendix C, and in Figures 5 and 6 below. Figure 5. “Reason” through Three Generations of Critical Theory ---------------------Figure 5 about here ---------------------- 25 Figure 6. Reduced to only the shared ways of describing “Reason” ---------------------Figure 6 about here ---------------------If we look at the center of Figure 5, we might be surprised to see that these “three generations” do not share all that much. Indeed, they share the important triad of reason, social, and nature. The exact argument varies from author to author, but Frankfurt philosophers have long found it impossible to analyze the concept of reason without referencing its intimate connection to ideas of nature and the social. For example, Horkheimer and Adorno (1947) discussed society’s power over nature and nature’s power over the mind. An individual’s reason, they argue, is exercised in the service of his natural interest; nature gives reason its “ends.” Human beings, however, become mere material as the whole of nature is subordinated to society; thus, those natural interests are actually socially determined. Habermas (1981) explained social action by contrasting it to non-social action: the use of instrumental reason to act upon and achieve goals with regard to nature. Social action, he argued, should not be instrumental but communicative and oriented toward reaching mutual understanding. More interesting, however, is how this triad of authors works dyadically. By examining the dyadic relations among authors, we might elucidate a process underlying Critical Theory’s establishment as an intellectual school: the bridging of cultural holes. Notice first of all a surprising finding: Horkheimer and Adorno and Habermas have nothing in common besides the basic reason-social-nature triad. Their cultural relation is—while arguably present—surprisingly weak. Indeed, Habermas expresses quite explicitly his debt to Horkheimer and Adorno’s theory of reason. They are, after all, two of the primary concepts in which he embeds his own ideas on 26 the topic. However, he does not accept Horkheimer and Adorno’s understanding of reason as inhering in the individual human self’s thought. He develops instead the idea of a communicative reason that constitutes action and goes on to contrast this reformulated concept with Weber’s notions of rationality and societal rationalization. The apparent ambiguity toward Horkheimer/Adorno may suggest that Habermas knows he is expected to engage with them when theorizing about reason—for an Adorno student, professor at Frankfurt, and recent recipient of the Adorno Prize in Philosophy, this was an obvious citation—but fundamental theoretical differences prevent specifics of their theories from converging. So although there is desire to connect with Horkheimer and Adorno, this does not come to fruition. There is essentially a cultural hole between these two generations. There is a social relationship as well as many other reasons to believe there should be a cultural relationship, but there is none. Instead, there is a moment of cultural discontinuity. By contrast, Honneth is heavily indebted to his predecessors, sharing considerable linkages with both. Like Habermas, his debt to Horkheimer and Adorno is explicit—they are also key concepts in which his discussion of reason is embedded. But unlike Habermas, this is justified by a strong theoretical connection. Honneth holds firmly to Horkheimer/Adorno’s original understanding of reason as inhering in human thought, and its situatedness in questions of power. And even the semantic relations that Honneth and Horkheimer/Adorno do not explicitly share demonstrate a common orientation toward understanding reason as a correlate of the capitalist economy. For Horkheimer/Adorno, it is intimately bound with bourgeois tendencies, for Honneth with questions of labor. Honneth has a likewise strong cultural relationship with Habermas. Both share an intellectual debt to Horkheimer/Adorno’s theory of reason. And their critiques are of specifically instrumental reason, a shared reference to Max 27 Weber. In addition, their works tend toward metacritique—more interested in examining the history of social philosophy and criticism, than critiquing the state of society as such. Where there was a gulf between Horkheimer/Adorno and Habermas, Honneth serves as a bridge. His work forges post hoc the theoretical unity that makes Critical Theory coherent. The pattern of semantic structural convergence we see here conforms to Honneth’s philosophical mission: to infuse the basic framework of Horkheimer/Adorno’s vision of a capitalist society that dominates nature and the individual with Habermas’s possibility for liberation through interpersonal communication and social action. This infusion could solve problems raised by the quasi-functionalist element of Horkheimer/Adorno’s theory that drastically limits the scope of individual creativity and curtails the possibility of liberation (Honneth 1995: 75). Honneth would eventually complete this first/second generation synthesis with his theory of recognition, for which he sows the seeds here in Critique of Power. An elaboration on the “reason” theme, the theory of recognition draws on Hegel’s dialectic of selfconsciousness to unite in a process the individual-reflective elements of Horkheimer/Adorno’s theory with Habermas’s understanding of reason inhering in communication. According to Honneth, the very possibility of communicative action depends on actors mutually recognizing each other as moral beings who are worthy of respect and capable of reason. I can only recognize my own reason when I recognize the reason in the other; the other can only recognize his own reason when he recognizes mine. Liberation depends on reflection and communication reinforcing one another. We might conclude, then, that the third generation brought this tradition together. And it did so by bridging the cultural hole formed by its predecessors. Thus, my argument supports two existing arguments about the Frankfurt School’s generational structure that seem to oppose one 28 another, showing how they are in fact compatible., There is a clear break between the first and second generations. It seems that in terms of the primary problem of “reason,” the second generation wanted to distance itself from the first—at least at the time the Theory of Communicative Action was written. But the school also develops in three cohesive generations because the third generation redrew these connections, unifying these philosophers as a school, a tradition that shows coherence through three generations. We had reason to expect this finding. Scholars within the Critical Theory tradition have long known that Habermas had a difficult relationship and little philosophically in common with his mentors. At the height of his career, he embraced currents of thought such as pragmatism and linguistic analysis to which Horkheimer and Adorno were hostile, building new approaches that seriously challenged Horkheimer and Adorno’s formulations (Honneth 1995: 86). And Honneth has been candid about his attempt to unite the disparate first and second generations in his third.14 This semantic relational method confirms this hypothesis, demonstrating its potential to approach questions in the future for which we have no ready answer. We can go on to ask whether cultural ties strengthen, weaken, or change in theoretical focus over time, or whether the school in general goes through periods of greater and lesser intellectual unity. The possibilities for discovering through an examination of multiple semantic meanings the longitudinal evolution of cultural ties, and with it processes of influence and institutional establishment, are endless. 6. Conclusion: Structures of Culture and Cultural Holes Semantic relational analysis offers exciting new possibilities for examining processes of cultural production. Because processes are complex and unfold over time, they are particularly demanding of sociological methods. Methods for studying them must contend at least somewhat 14 Honneth in fact explained to me his attempt to do so in person at his Frankfurt office. 29 with the inherent complexity of cultural phenomena; they must be capable of incorporating the large amount of data necessary for longitudinal analysis; finally, they must generate cultural data that can be integrated fluidly with sociological data. These are precisely semantic relational analysis’s strengths. Here, it was first shown to accomplish a rigorous and detailed structural treatment of cultural phenomena. Computer-assisted formal semantic analysis was able to schematize the semantic meaning of Habermas’s Theory of Communicative Action. It captured how Habermas constructed his main argument through engagement with various interlocutors. The schematization produced resonates with accepted general interpretations of the text, successfully identifying the main themes that most readers would understand as central to it. Thomas McCarthy’s introduction to Theory of Communicative Action (1984) and David Ingram’s Habermas: Introduction and Analysis (1989) are two of the most authoritative general discussions of the text’s themes. And consonant with my formalization, both authors emphasize Habermas’s attempt to construct a theory of society that includes the lifeworld and system paradigms (McCarthy, 1984: xxvii; Ingram, 1989: 115), to critique of how life is penetrated by formal systems (1984: xxviii; 1989: 55), to critique of the instrumental rationality that pervades life rationalized by these formal systems (1984: xxxii, 1989: 64), to elucidate a different type of rationality in communication (McCarthy, 1984: ix-x; Ingram, 1989: 20), and to locate the potential for liberation through such rational communication (1984: xxxvii, 1989: 180). Thus, semantic relational analysis appears a reliable technique that may be applied to texts for which we do not have existing interpretations. In highlighting the similarities between the results of formal analysis and theoretical exegeses, however, it is important to note that comparing the two is of limited use. Semantic analysis and theoretical exegeses have entirely different goals. So while a comparison can 30 confirm the internal validity of a semantic method, semantic analysis does not otherwise attempt to intervene into philosophical discourse. Formal semantic analysis is singularly devoted to a text, creating a summary of it in the form of a concept map. It unites and structures key concepts, regardless of how difficult it might be to incorporate them into a single coherent theoretical synthesis. By contrast, most theoretical exegeses are singular and interested. Their goal is not to summarize fully what an author wrote but to engage with a portion of the text to construct a new synthesis—a new idea—to further philosophical scholarship. As a result, they are partial by design; text not relevant to the construction of the new idea is excluded. Thus, when developing a formal semantic method, it may be useful to check its results against a few exegeses that are entirely or mostly devoted to identifying the main themes of the text, but the results are not meant to engage meaningfully with most other interpretations. Still, semantic analysis can make important contributions to philosophical discourse by allowing researchers to compare texts’ main themes in a disinterested way. This lends fresh perspective to questions like how authors fit together as a school. Most researchers who analyze the development and continuity of a philosophical school find at least some of its core ideas personally compelling; thus something is at stake when they define that school. The fact that their analyses are perhaps necessarily tendentious in no way means that they are necessarily unsound, but we must recognize that, at an extreme, such analyses may turn out to be selffulfilling prophecies (think of Alexander’s (1990) successful construction of a Durkheimian sociology that included him as the current representative, contrasted to Collins’s (1986) unsuccessful and temporary attempt to make a Weberian sociology that emphasized geopolitical concerns). Secondary literature on the Frankfurt School (e.g. Jay 1973, Kellner 1975, Wiggershaus 1995, Türcke & Bolte 1994, Honneth 2007, Bonner 2011, Held 1980, Wheatland 31 2009) is no exception. There may be a tendency to define the school first in a way that reflects one’s interests, and deduce from that definition how authors’ ideas converge to form that school. For example, Jay (1973), having defined the school as a certain form of Left Hegelian neoMarxism, includes in the school members of the Frankfurt Institut für Sozialforschung only after their emigration to the United States. Honneth (2007) defines the school as three cohesive generations with himself at its helm. We may, however, want the option of handling questions regarding the convergence of different thinkers’ conceptual schemes that are divorced, or at least divorceable, from the arguments about how a school should be defined. Semantic relational analysis offers this alternative approach. It looks empirically at how authors’ ideas overlap. Then we may or may not choose to induce from that pattern of convergence a definition of the school. Here, the similarities and differences in semantic patterns among three authors— Horkheimer/Adorno, Habermas, and Honneth—were analyzed, revealing strong, weak, and qualitatively different cultural ties between these school members. These patterns, when analyzed in tandem with social structural data, demonstrate how the Frankfurt School of Critical Theory was forged by the school’s third generation. It united the tradition by bridging the cultural hole formed by its predecessors with a particular conception of “reason.” Analyses like this can be scaled up easily. Incorporating works by other authors within the same time frame or by the same authors in an extended time frame could add even more insight into the development of philosophical schools. Semantic relational analysis provides a strong empirical foundation for studying cultural relations in general and may therefore help us understand other processes at work in cultural institutions. Influence, for example, might be represented by existing social relationships 32 accompanied by cultural attributes converging over time. The convergence can be measured as strong or weak, sudden or gradual, unidirectional or bidirectional. We can raise sociological hypotheses about whether certain types of social relations tend to correlate with certain types of influence. Perhaps authority relationships tend to yield strong unidirectional influence; perhaps close friendships tend to exhibit strong, bidirectional influence. The tendency of people to become attracted to culturally similar others might be represented by cultural ties preceding the formation of social ties. From there, we can answer questions as to the types of cultural ties that have certain social structural repercussions. Maybe there are limits to the structuring potential of similarity, where cultural convergence that is too strong becomes socially repulsive instead of attractive. Perhaps strong friendships correlate to a wide array of weakly converging tastes and beliefs, whereas acquaintanceships correlate to a very small set of strong convergences. Perhaps having too many common beliefs ends up—counterintuitively—tearing close relationships apart over time. These are issues at the core of understanding social relationships, as well as understanding the role of culture in social life. While it has long remained a challenge to study them systematically, the semantic relational approach presents some exciting new possibilities. 33 Works Cited Alexander, Jeffrey. 1990. Durkheimian Sociology: Cultural Studies. Cambridge, UK: Cambridge University Press. Anderson, Joel. 2000. “The ‘Third’ Generation Frankfurt School.” Intellectual History Newsletter 22. Baldassari, Delia and Mario Diani. 2007. “The Integrative Power of Civic Networks.” The American Journal of Sociology 113: 735–80. Bearman, Peter S and Katherine Stovel. 2000. “Becoming a Nazi: Models for Narrative Networks.” Poetics 27:69-90 Bolte, Gerhard, ed. 1989. Unkritische Theorie: Gegen Habermas. Lüneburg: zu Klampen. Bonner, Stephen. 2011. Critical Theory: A Very Short Introduction. Oxford, UK: Oxford University Press. Breiger, Ronald L. 1974. “The duality of persons and groups.” Social Forces 53: 181-90. ______________. 2000. “A Tool Kit for Practice Theory.” Poetics 27: 91-115. Burt, Ronald. 1987. "Social Contagion and Innovation: Cohesion Versus Structural Equivalence." Ronald S. Burt; American Journal of Sociology 92(6), pp. 1287-335. Carley, Kathleen. 1993. "Coding Choices for Textual Analysis: A Comparison of Content Analysis and Map Analysis, " in Peter Marsden, ed., Sociological Methodology Vol. 4. Oxford: Blackwell. _______________. 1994. “Extracting Culture Through Textual Analysis.” Poetics 22:291-312. _______________. 1997. “Network Text Analysis: The Network Position of Concepts.” in Text Analysis for the Social Sciences: Methods for Drawing Statistical Inferences from Texts and Transcripts, edited by Carl Roberts. Mahwah, NJ: Lawrence Erlbaum Associates. Carley, Kathleen M. et. al. 2013. Automap User's Guide 2013. Institute for Software Research, Carnegie Mellon University. Carley, Kathleen M. and Michael Palmquist. 1992. "Extracting, Representing, and Analyzing Mental Models." Social Forces, 70: 3: 601-636. Cerulo, Karen A. and Ruane, Janet M. 1998. “Coming Together: New Taxonomies for the Analysis of Social Relations.” Sociological Inquiry 68: 3: 398-425. Coleman, James S., Elihu Katz and Herbert Mendel. 1957. “The Diffusion of an Innovation among Physicians.” Sociometry, 20: 253-270 Collins, Randall. 1986. Weberian Sociological Theory. Cambridge, UK: Cambridge University Press. della Porta, Donatella. 1988. “Recruitment Processes in Clandestine Political Organizations: Italian Left-Wing Terrorism,” in Bert Klandermans, Hans Kriesi, and Sidney Tarrow (eds), From Structure to Action. Greenwich: JAI Press. Dilthey, Wilhelm. [1883] 1990. Einleitung in die Geisteswissenschaften: Versuch einer Grundlegung für das Studium der Gesellschaft und der Geschichte in Wilhelm Dilthey Gesammelte Schriften, Bd. 1. Göttingen: Vandenhoeck & Ruprecht. Evans, James. 2010. “Industry Induces Academic Science to Know Less about More.” The American Journal of Sociology 116:389-452. Fan, D. 1988. Predictions of Public Opinion From the Mass Media: Computer Content Analysis and Mathematical Modeling. New York: Greenwood Press. Franzosi, Roberto. 1989. “From Words to Numbers: A Generalized and Linguistics-Based Coding Procedure for Collecting Textual Data.” Sociological Methodology 19:263-298. 34 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” ______________. 1994. “From Words to Numbers: A Set Theory Framework for the Collection, Organization, and Analysis of Narrative Data.” Sociological Methodology 24:105-136. ______________. 1997. "Mobilization and Counter-mobilization Processes: From the 'Red Years' (1919-20) to the 'Black Years' (1921-22) in Italy." Theory & Society, 26: 275-304. ______________. 1998. "Narrative Analysis and Why (and How) Sociologists Should be Interested in Narrative." Annual Review of Sociology, 24: 517-554. ______________. 2004. From Words to Numbers: Narrative, Data, and Social Science. Cambridge, UK: Cambridge University Press. Friedkin, Noah. 1998. A Structural Theory of Influence. Cambridge, UK: Cambridge University Press. Griswold, Wendy. 1987. "A Methodological Framework for the Sociology of Culture." Sociological Methodology. 17: 1-35. Habermas, Juergen. [1962] 1991. The Structural Transformation of the Public Sphere. Cambridge, MA: MIT Press. _______________. 1981. The Theory of Communicative Action. Boston: Beacon Press. _______________. 1985. Die neue Unübersichtlichkeit. Frankfurt: Suhrkamp, Held, David. 1980. Introduction to Critical Theory: Horkheimer to Habermas. Berkeley and Los Angeles: University of California Press. Honneth, Axel. [1989] 1991. Critique of Power: Reflective Stages in a Critical Social Theory. Cambridge, MA: MIT Press. ___________. 1995. The Fragmented World of the Social. Albany, NY: SUNY Press. ___________. [1992] 1996. The Struggle for Recognition. Cambridge, MA: MIT Press. ___________. 2007. Pathologien der Vernunft: Geschichte und Gegenwart der kritischen Theorie. Frankfurt: Suhrkamp. Ignatow, Gabriel. 2004. “Speaking Together, Thinking Together? Exploring Metaphor and Cognition in a Shipyard Union Dispute.” Sociological Forum 19(3): 405-433. Ingram, David. 1989. Habermas and the Dialectic of Reason. New Haven: Yale University \ Press. Jay, Martin. 1973. The Dialectical Imagination. London: Heinemann. Katz, Elihu, and Paul F Lazarsfeld. 1955. Personal Influence. Glencoe, Ill.: Free Press. Kellner, Douglas. 1975. " The Frankfurt School Revisited: A Critique of Martin Jay's The Dialectical Imagination." New German Critique, No. 4: 131-152. Kim, Hyojoung and Peter S. Bearman. 1997. “Who Counts in Collective Action? The Structure and Dynamics of Movement Participation.” American Sociological Review 62:70-93. Krippendorff, Klaus. 2004. "Reliability in Content Analysis: Some Common Misconceptions and Recommendations." Human Communication Research 30, 3: 411-433. Lieberman et. al. 2007. “Quantifying the Evolutionary Dynamics of Language.” Nature 449. Lizardo. Omar. 2006. "How cultural tastes shape personal networks." American Sociological Review 71: 778-807. Martin, John Levi. 2000. "What do Animals do all Day?" Poetics 27: 195-231. McAdam, Doug, and Ronnelle Paulsen. 1993. “Specifying the Relationship between Social Ties and Activism.” American Journal of Sociology 99:640-67. McCarthy, Thomas. 1984. Translator's Introduction to The Theory of Communicative Action. Boston: Beacon Press. Michel, Jean-Baptiste et al. 2010. “Quantitative Analysis of Culture Using Millions of Digitized 35 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” Books.” Science, December 16. Mische, Ann. 2003. “Cross-talk in Movements: Rethinking the Culture-Network Link.” Pp.258-80 in Social Movements and Networks: Relational Approaches to Collective Action, edited by Mario Diani and Doug McAdam. Oxford and New York: Oxford University Press. __________. 2007. Partisan Publics: Communication and Contention across Brazilian Youth Activist Networks. Princeton NJ: Princeton University Press. Mohr, John. 1994. "Soldiers, Mothers, Tramps, and Others: Discourse Roles in the 1907 New York City Charity Directory." Poetics 22: 327-357. __________. 1998. "Measuring Meaning Structures." Annual Review of Sociology. 24:345-370. Mohr, John and Vincent Duquenne. 1997. "The Duality of Culture and Practice: Poverty Relief in New York City, 1888-1917." Theory and Society 26:305-356. Mohr, John and Helene Lee. 2000. "From Affirmative Action to Outreach: Discourse Shifts at the University of California." Poetics 28(1):47-71. Mohr, John, Michael Bourgeois, and Vincent Duquenne. 2004. "The Logic of Opportunity: A Formal Analysis of the University of California's Outreach and Diversity Discourse." Center for Studies in Higher Education, UC Berkeley, Research and Occasional Papers Series. Moretti, Franco. 2005. Maps, Graphs, Trees: Abstract Models for Literary History. London and New York: Verso Moretti, Franco. 2009. “Style, Inc. Reflections on Seven Thousand Titles.” Critical Inquiry 36. Morris, Charles. 1938. Foundations of the Theory of Signs. Chicago: University of Chicago Press. Namenwirth, J. and R. Weber. 1987. Dynamics of Culture. Boston: Allen & Unwin. Pachucki, Mark A. and R.L. Breiger. Forthcoming. “Cultural Holes: Beyond Relationality in Social Networks and Culture.” Annual Review of Sociology 36 (2010). Peirce, Charles Saunders. [1940] 1955. “Logic as Semiotic: The Theory of Signs” in Philosophical Writings of Peirce (pp. 98-121). New York: Dover. Postone, Moishe. 2009. History and Heteronomy: Critical Essays. Tokyo: University of Tokyo Center for Philosophy Publications. Schultz, Jennifer and Ronald Breiger. 2010. “The Strength of Weak Culture.” Poetics 38(6). Shively, JoEllen. 1992. “Cowboys and Indians: Perceptions of Western Films Among American Indians and Anglos.” American Sociological Review. Vol. 57, No. 6: 725-734. Steinert, Heinz. 2007. Das Verhängnis der Gesellschaft und das Glück der Erkenntnis: Dialektik der Aufklärung als Forschungsprogramm. Münster: Westfälisches Dampfboot. Stone, Philip J. et al. 1966. The General Inquirer: A Computer Approach to Content Analysis. Cambridge, MA: MIT Press. Sullivan, M. 1973. Perceptions of Symbols in Foreign Policy. Ann Arbor: Inter-University Consortium for Political Research. Teil, Genevieve and Bruno Latour. 1995. “Can Association Networks do more than Formal Rules?” Stanford Humanities Review 4(2).’ Türcke, Christoph and Gerhard Bolte. Einführung in die Kritische Theorie. Darmstadt: Wissenschaftliche Buchgesellschaft. Vaisey, Stephen and Omar Lizardo. 2009. “Can Cultural Worldviews Influence Network 36 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” Composition?” Social Forces 88(4). Wheatland, Thomas. 2009. The Frankfurt School in Exile. Minneapolis: University of Minnesota Press. Wiggershaus, Rolf. 1995. The Frankfurt School: Its History, Theories, and Political Significance. Cambridge, MA: MIT Press. Zerubavel, Eviatar. 2012. Ancestors & Relatives: Genealogy, Identity, & Community. New York: Oxford University Press. 37 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” APPENDIX Appendix A: Testing Semantic Neighborhoods A.1 Clause, Sentence, Paragraph The first grammatical unit examined is the clause. These are strings of text set apart with commas, semicolons, and periods.15 Averaging 11 words in length in Theory of Communicative Action, the clause is the smallest grammatical unit examined here. One could argue that each clause represents an idea. For example, in delineating the following clauses “In the analytic dialogue the roles are asymmetrically distributed; the analyst and the patient do not behave like proponent and opponent” (Habermas 1981: 21), the semicolon serves the desired purpose. It separates a statement about analytic dialogue from a statement about an analyst and a patient, both of which are coherent expressions. But the method will also make “mistakes,” eliminating words from clauses because they are in comma-separated lists. For instance, commas break up the single coherent expression “Someone who explains his libidinous reaction to rotten apples by referring to the ‘infatuating,’ ‘unfathomable,’ ‘vertiginous” smell, or who explains his panicked reaction to open spaces by their ‘crippling,’ ‘leaden,’ ‘sucking’ emptiness, will scarcely meet with understanding in the everyday contexts of most cultures” (Habermas 1981: 17) into seven different clauses. False negatives—semantic relations that should be identified but are not— abound. On the other hand, we can be fairly certain that words often co-present in these small lexical units (>100 times in Theory of Communicative Action) are truly “related” in a text. So while clauses will produce many false negatives, they will produce very few false positives. Co-presence in sentences has been suggested by Genevieve Teil and Bruno Latour (1995) as the best means for semantic-associational analysis. They argue that the simple co-presence of 15 These lexical units are simply strings of text set apart from each other with punctuation; As such, not every string is a true grammatical clause. However, the punctuation-split operation closely resembles breaking up the text into grammatical clauses, so I will refer to these units as “clauses” for shorthand. 38 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” concepts in sentences provides surprisingly rich meaning. With an average length of 28 words, the sentence is the mid-sized unit examined here. While false positives should remain relatively low, the potential for false negatives remains. One can imagine that in texts, especially ones that discuss complex ideas, a coherent idea requires more than a single sentence to express. For instance, the sentence cited above about the dialogue between an analyst and a patient certainly expresses a coherent thought, but it does not capture the entire larger idea that helps Habermas build his fundamental argument. The sentence is actually embedded in a statement about selfdeception and irrationality that takes place over a few sentences. The last grammatical unit examined is the paragraph. The paragraph is the largest unit examined here, averaging around 108 words. The paragraph level offers great advantages when a text is long. Even more than sentences, we can imagine paragraphs as the building blocks for a complex system of meaning. In fact, this is how we are taught to use the English language. Each paragraph represents a main idea. If one reads through the topic sentences of a wellconstructed text, one should get the skeletal structure of its conceptual arc. Of course, paragraphs are still not perfect. Semantic relations according to paragraph co-presence bridge greater distance. Thus, these larger units tend toward true negatives, but false positives. A complete idea may require more than one sentence to express—and a paragraph method will capture that—but a poor writer might still express several unrelated ideas in a single paragraph. I have tested these suspicions by analyzing Habermas’s text these three different ways. The 30 strongest semantic relations for each approach are listed in Table A.1; network visualizations are provided in Figures A.1, A.2, & A.3. 39 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” Table A.1: 30 top relations—clause, sentence, paragraph -------------------------Table A.1 about here -------------------------Figure A.1: Habermas by Clauses -------------------------Figure A.1 about here -------------------------Figure A.2: Habermas by Sentences -------------------------Figure A.2 about here -------------------------If we first compare the results for the clause analysis with those of the sentence analysis, we see that from clause to sentence seven out of thirty relations— mutual-understanding, understanding oriented, societal-rationalization, theory-parsons, everyday-practice, modernsocieties, and system-integration—have been replaced with action-cultural, world-action, lawlegal, action-way, action-weber, action-parsons, action-knowledge. The move from clause to sentence exhibits four patterns of note: 1) In almost every case a simple bigram has been replaced with a semantic relation bridging greater distance. 2) The graph produced becomes a more coherent structure; semantic cliques previously separate from the main structure become integrated into it. 3) Semantic “hubs” such as action become increasingly evident. 4) The frequency of relations increases substantially. These four patterns can be considered 40 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” advantageous because they move the structure toward greater coherence, reliability, and intricacy. But with these advantages comes at least one drawback: dominant nodes become amplified, possibly obscuring other structural patterns. Here, action dominates, swamping the other nodes. Connections to action compose the primary—if not only—important structural feature, making the graph somewhat less informative. Figure A.3: Habermas by Paragraphs -------------------------Figure A.3 about here --------------------------When we move from sentence to paragraph, the same patterns persist. Everything “gained” by moving from clause to sentence has been retained in the paragraph analysis. On top of that, seven other relations—understanding-reaching, social-integration, action-oriented, world-objective, understanding-world, action-knowledge, and systems-theory are replaced with social-lifeworld, validity-action, action-means, actor-action, action-society, social-society, and action-oriented. Again, bigrams are replaced by relations of greater distance, the graph has become more coherent, semantic hubs become clearer, and the sheer frequency of relations increases substantially, but action swamps the structure. Still, because of the connectivity and frequency of relations in the relational matrix produced—if not the slightly sounder logic underlying it—the paragraph method appears at present to produce the most useful data for further analyses. A.2 Analysis of N-Grams 41 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” A window of size n-words moves over the entire text, tallying the number of times words are co-present with one another. N-gram weighting is done by allowing a relation to count in every text window in which it is present, rather than only once in each analysis. For example, analyzing 3-grams in the sentence “The quick brown fox jumped over the lazy dog,” brown-fox would count twice in the analysis, whereas brown-jumps would count once. Analyzing 4-grams, brown-fox would count three times, brown-jumps would count twice, and brown-over would count once. Thus, relations that bridge greater textual distance must be substantially more frequent than ones that are textually close to prove their validity and outweigh close ones in the analysis. The results of the n-gram analysis are summarized below in Figure A.4. On the left is a list of semantic relations; on the top is a listing of n-gram sizes. Presence of the bar indicates that the semantic relation is identified by the n-gram. Figure A.4. N-grams of Size 3-300 -------------------------Figure A.3 about here --------------------------This crucial point is at n = 104-146. We can see relations disappear above, and appear below power-labor as n300. From n of size 3-40, the results of the analysis change rapidly. Bigrams like speech act and symbolically mediated are replaced by semantic relations that bridge greater textual distance, but are substantially more frequent. But as the n-gram grows, changes to it become scarcer and scarcer. When it gets larger than n = 104, the relational matrix largely stabilizes. From n = 104-146 until n = 300, the set of relations remains essentially the same, with only the 30th (weakest) relation replaced at n = 147 (social-integration social-society), n = 161 42 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” (social-society validity-action), and n = 211 (validity-action system-parsons). Nothing that enters the relational matrix after n = 104 moves above the lowest (30th) position. This means that the weighting of “closer” relations begins to win out. After n = 104, no new semantic relations are frequent enough to outweigh the closer ones already included in the relational matrix. It is therefore evident that n = 104-146 is the critical moment in the analysis; it is the point where group size is maximized, but the semantic relations remain meaningful. Figure A.5 is the graph of the semantic network at n = 104-146. Figure A.5: Habermas Semantic Network at n = 104-146 -------------------------Figure A.5 about here --------------------------This key n-gram is closest in size and results to a paragraph (recall that the average paragraph length is 108 words). This n-gram is also the most similar to a paragraph in terms of its results, with only two low-ranking semantic relations differentiating the two matrices: validity-action and actor-action in the paragraph matrix turn into social-integration and reaching-understanding in the 104-gram. This is likely the result of the weighting mechanism used in the n-gram analysis, as social-integration and reaching-understanding are textually close relations (one a bigram, the other practically a bigram) that are used fairly often. Appendix B: Analyzing Texts in Translation It is tempting to assume it is always preferable to perform semantic analyses in a text’s original language, but this will not always yield the most reliable results. Whether it does or not depends on characteristics of both languages as they bear on the type of semantic analysis to be 43 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” performed. A semantic relational analysis based on co-presence in semantic neighborhoods is most reliable when each concept’s symbolic representation is consistent throughout a text; that is, a concept is expressed in as few forms as possible—concept A is represented by only A, not by A as well as Ar, An, and As. Due to its conjugation and declension rules, as well as the tendency to create compounds, concepts are relatively inconsistent in German. We are therefore likely to achieve more reliable results when analyzing texts in English, given that the text was translated with reasonable care. Semantic analysis in German is symbolically inconsistent because words can take various different “forms” depending on grammatical context. German verbs are conjugated with great variability and adjectives are declined according to case, number, and gender, whereas these words seldom morph in English. Verbs vary more in German than in English. For example, in English, I consider, you consider, they consider, we consider, and he/she considers. Conjugation varies the verb “to consider” only modestly, and to automatically code a verb as including its third person singular form (original form + “s”) is computationally simple. By contrast, in German, I überlege, you überlegst, he/she überlegt, we überlegen, and they überlegen—and these are only the present and regular forms of the verb; other tenses and constructions require further conjugations. Adjectives are declined according to case, number, and gender. Something that is “loud” and written only as loud in English may appear in German as laut, laute, lauter, laute, lautes depending on its grammatical case, whether or not a definite article is used, and characteristics of the modified noun. Indeed, natural language processing can solve these problems through an automated identification of verbs or adjectives, and a programmed grouping of verbs and adjectives with common stems, regardless of their endings. However, each manual “fix” of this sort in the code is both time consuming and a potential 44 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” source of error for an automated analysis. Algorithms for part of speech identification are imperfect and will almost certainly fail to identify some verbs or adjectives, making the results of even an adjusted German analysis at least somewhat less reliable than those of an English analysis. An even more difficult source of symbolic inconsistency in the German language is its tendency to compound words. There is a tendency to both combine two or more words (usually nouns) into a single one, as well as to separate verbs’ prefixes from their bases in certain grammatical constructions. First, two individual nouns may be combined into a single compound noun. We can take for example the concept of Produktion (production, as in manufacturing), which as a cornerstone of Marxist thought, is discussed often in Critical Theory. Nouns that contextualize or modify Produktion are sometimes united with it to create a compound noun. For example, “forms of production” is normally expressed as Produktionsformen, “process of production” as Produktionsprozess, “relations of production” as Produktionsverhältnisse. Now, if nouns and their modifiers were always compounded, the analysis could adjust for this pattern. But unfortunately, compounding is not always the norm. For example, “cessation of production” is normally expressed as Einstellung der Produktion, “modernization of production” as Modernisierung der Produktion. Also, only nouns compound; other modifiers such as adjectives are rarely, if ever, compounded with a base noun. For instance, “limited production” and “efficient production” are expressed as begrenzte Produktion and effiziente Produktion, respectively. What is more, regardless of whether compounding a phrase is the norm, it remains optional; the writer or speaker does it largely according to her own preference. Altogether, this tendency to compound nouns makes for significant noise in any semantic analysis because some instances of a noun and its modifiers would be expressed as 45 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” single compound words, whereas other instances as various unique words. Second, verbs are often “separable” (trennbare) compounds that are divided into prefixes and bases upon conjugation. This creates the same problem as creating compound nouns: sometimes two words will be attached, sometimes detached, so a computational analysis will treat these two forms differently even though they represent the same concept. Still more problematic is that with verbs, the prefix and the base together as one word can have a meaning that is quite distinct from its two elements as independent words. For example, we can think of the word unterbringen, which as a compound means “to house or accommodate” a guest. In many grammatical constructions, the verb is separated: “I house the guest” is expressed as ich bringe den Gast unter. But the two fragments, unter and bringen are words in themselves with independent meanings, literally “under” and “to bring.” If analyzed computationally many instances of accommodation would be read erroneously as “bringing somebody under,” indicating something that sounds more like an act of murder than an act of hospitality. And adjusting a computer program to deal with compounding may be much more difficult than to deal with conjugation and declension, since it is relatively unregulated. The symbolic stability of English would therefore facilitate a more reliable semantic analysis. Analyzing texts in English is, however, only advantageous if the text’s translation is high quality. Indeed, there is no such thing as a perfect translation since words in different languages cannot be exchanged in a 1:1 relationship, but philosophical translations may be some of the best available. They are often accomplished as the result of intense contestation among scholars in the philosophy being translated (as opposed to being the work of professional translators) and must withstand a long process of painstaking review and criticism before publication. Significant difficulties in translation are normally noted in the translator’s note of a volume, 46 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” alerting a semantic analyst to adjust for them. Most importantly, concepts of philosophical import are normally understood as “technical terms;” they are therefore treated with the utmost care and usually kept consistent between the original text and the translation. There is, however, one common translation practice that is problematic for an analysis of semantic relations based on co-presence in semantic neighborhoods: English translations often add paragraph breaks not present in original German texts. This is because German paragraphs can be unseemly in length, cobbling together several distinct thoughts (although this is more the case for pre-20th centrury texts than for contemporary ones). Depending on how often paragraphs are broken up in a given text (and prevalence will vary widely), there may be a significant tension between the results of an English and a German analysis. But even in this case, the argument may fall in favor of using the English version. Paragraphs are broken up in translation to better organize the text, to make it more understandable when the original writing style is unsystematic. If a translation makes for qualitatively better semantic neighborhoods—paragraphs that truly represent a single idea—by re-applying the rules of paragraph breaks, the translation may prove better for semantic relational analysis. Again, whether this is at all a concern for a study depends on the text in question, but it is important to consider paragraph beaks when analyzing translations. Altogether, we still have reason to believe that the distribution of philosophical concepts is largely the same over the course of both versions. Performing semantic relational analysis on the translation therefore seems reasonable (although this may not hold for other approaches to semantic analysis). Even though a small portion of a text’s semantic structure may be altered by translation, the costs of remaining in German outweigh the benefits. Still, although analyzing in English is generally simpler and more reliable, analyzing in German reliably is a worthy goal—if not only because not all German texts are translated into 47 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” English well. The first steps toward doing so are to adjust the programmed analysis for verb conjugation and adjective declension, as well as commonly compounded/separated words. This should improve the reliability of German language analyses. Appendix C: Binarized Semantic-Relational Matrices to Discover Convergence in Meaning Structures Figure C.1: Horkheimer/Adorno on “Reason” Matrix -------------------------Figure C.1 about here --------------------------Figure C.2: Habermas on “Reason” Matrix -------------------------Figure C.2 about here --------------------------Figure C.3: Honneth on “Reason” Matrix -------------------------Figure C.3 about here --------------------------Figure C.4: Semantic Relational Convergences (Marked on Honneth Matrix) -------------------------Figure C.4 about here --------------------------- 48 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” FIGURES Figure 1. The Structure of Communicative Action Figure 2. Chapter 1: Introduction: Approaches to the Problem of Rationality 49 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” Figure 3. Chapter 6: Intermediate Reflections: System and Lifeworld Figure 4. Chapter 5: The Paradigm Shift in Mead and Durkheim: From Purposive Activity to Communicative Action 50 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” Figure 5. “Reason” through Three Generations of Critical Theory 51 Monica Lee – Writing Sample #2 – “Structuring Communicative Action” Figure 6. Reduced to only the shared ways of describing “Reason” 52 Table A.1: 30 top relations—clause, sentence, paragraph Clause Relations Times occurred Sentence Relations Times occurred Paragraph Relations Times occurred communicative-action* validity-claims* action-theory speech-acts* understanding-reaching social-integration* action-systems action-orientations action-system understanding-action social-action social-system* action-oriented action-concept system-lifeworld social-theory* social-world systems-theory* rationalization-weber world-objective mutual-understanding* rationality-action understanding-oriented action-lifeworld understanding-world societal-rationalization* theory-parsons everyday-practice* modern-societies* system-integration* 340 269 234 230 221 209 176 174 170 169 168 158 152 147 142 131 123 123 121 121 118 118 117 116 105 101 90 89 88 88 communicative-action* social-action action-theory speech-acts* validity-claims* action-system understanding-action action-lifeworld action-orientations* social-system* understanding-reaching action-systems social-integration* action-oriented action-concept system-lifeworld social-world rationality-action social-theory* world-objective rationalization-weber action-cultural understanding-world world-action law-legal action-way action-weber action-parsons action-knowledge systems-theory* 466 378 362 344 334 314 302 273 268 268 256 255 252 245 233 226 225 212 206 203 199 188 182 175 173 170 167 166 160 159 social-action action-system action-theory communicative-action* social-system* understanding-action action-lifeworld action-concept law-legal action-weber action-parsons social-theory* social-lifeworld rationality-action action-cultural action-systems social-world system-lifeworld world-action validity-claims* rationalization-weber action-way validity-action speech-acts* action-means action-orientations actor-action action-society social-society action-oriented 1730 1229 1210 1209 1011 1006 961 917 836 788 784 777 753 723 703 698 670 668 667 659 657 636 628 623 600 588 581 560 555 538 *indicates that the relation may be considered a bigram 53 Figure A.1: Habermas by Clauses Figure A.2: Habermas by Sentences 54 Figure A.3: Habermas by Paragraphs 55 Figure A.4. N-grams of Size 3-300 56 Figure A.5: Habermas Semantic Network at n = 104-146 Figure C.1: Horkheimer/Adorno on “Reason” Matrix 57 Figure C.2: Habermas on “Reason” Matrix Figure C.3: Honneth on “Reason” Matrix 58 Figure C.4: Semantic Relational Convergences (Marked on Honneth Matrix) 59