Multiple Systems and Multiple Time Scales of Language Dynamics: Coping with Complexity

formal structure • Independent of dynamics (grounding problem) • The causal role for dynamics (physical events) difficult to explain • Unclear how symbols get into a cognitive system • Independent level of explanation (res cogitans) • Symbols are manipulated according to syntactic rules, independently of their meaning (on the basis of their shape alone). Physical structure • Structure which arose in the context of particular system-organism dynamics • As a physical structure it can have a causal role with respect to dynamical events. • The presence of a particular structure in a particular time and place is not accidental: It is an outcome of a history of a structure within the system (including natural selection). • Not an independent level of explanation (res cogitans always with res extensa). • Symbols do not “behave” according to rules but rather their structures depend on dynamics within which they arose and which they control. Multiple Systems and Multiple Time-scales of Language Dynamics 43 Language as a System of Replicable Constraints Applied to language, this third way, (i.e., one that does not try to dissolve symbols in dynamics nor reduce dynamics to symbols but advocates complementarity of description), renders a particular picture of language, in which symbols are understood as described above. This picture diverges both from the common intuitive folk knowledge about language and the one present in the traditional theories (such as generativist theory). I will briefly (and in a simplified way) describe these common assumptions and claims that diverge the most from the theory of language as a system of replicable constraints, and point to the ways they might have skewed our thinking about language and prevented seeing more fruitful ways to study it, which are, in turn, enabled by the alternative view. Forms of Language Do Not Map onto Meanings Most theories of language accept that language is a system of formal symbols that are about something due to a mapping relation between form and meaning. Symbols of natural language refer to things in the world or ideas in the head. Therefore linguistic processing is about something (has meaning) because at any moment one can substitute this something for a given symbol, or map the content onto a symbol. Alternative view: If symbols are understood as constraints on dynamics, then it is not possible that the relation between a symbolic form and its meaning is the one of mapping. Mapping is a relation between two sets of identifiable elements. The relation of constraining does not offer this feature (identifiability of elements) on the side of symbol’s action. Symbolic forms are distinguishable entities, but the dynamics they constrain is different every time the constraining happens. Systems in which the symbolic forms are immersed are complex and dynamical: The effects of a symbol’s action may be vastly different depending on the state of the system at a given moment and may change non-linearly depending on a number of other variables impinging on the system. Think about a peg in a flow of fluid: It will disturb the flow, creating vortices. The vortices will be a bit different each time they form, and vastly different if the speed of the flow and the viscosity of the fluid changes. Symbols are structures that substitute (map onto) something else. Semantics of symbols is in the mapping relation between a set of forms and a set of meanings (traits or referents) Symbols are structures that control dynamics. It is difficult to predict the outcome of their action (meaning). Meaning is always underspecified which means that mapping relation is not possible (or can only be an approximation) Mapping or referring relation is a binary (0 or 1) relation Constraints on dynamics may have different strengths (leaving different role to contexts) Table 1: Symbol: Redefinition (Continued) Symbol in a Formal System Symbol as a Constraint 44 Joanna Rączaszek-Leonardi However if mapping is not a good way to characterize the relation between form and meaning then substituting cannot be the process that gives meaning to symbols. The proposed constraining relation, which links symbols to their meaning, is much more complex than mapping. First, symbols work as constraints not only in an individual here and now. Constraining happens also with respect to dynamics of a conversing dyad, or other interactions (e.g., Jennings & Thompson, 2012; Cowley, 2011) and on time scales longer than the on-line scale (most notably in development). Second, unlike the relation of mapping, or referring, the relation of constraining is not all-or-none. Constraining can be strong, for example, almost unequivocally and independently of context specifying the referent (e.g., Barak Obama) or can be weak, constraining only some aspects of potential meaning, leaving a lot for the individual and contextual dynamics (e.g., deictic here). The Structure of Language Is Not Exclusively Due to the Individual Cognitive Machinery In the approach that informed the last several decades of psycholinguistics, language is treated, above all, as an individual cognitive skill: Individuals possess cognitive machinery, which generates all grammatical and none of the ungrammatical sentences of one’s native language (save the performance constraints). Alternative view: It is true that individuals are the source of the physical stimuli that have constraining power. Yet their production is crucially dependent on: i) being in a certain situation here and now, and ii) being prepared, or tuned for linguistic and—more broadly—social interaction both in evolution, acculturation and development. From this perspective language becomes, above all, a way of physical and cognitive coordination. The primacy of the individual in explanations may obscure the natural collectivity, the primacy of we (see Rączaszek-Leonardi & Cowley, 2012) that lies at the basis of communion or synergy formation. In other words, rather than being a machine in individual’s head, language is seen as a system of constraints for controlling collectivity, providing means for flexible coordination of elements into functional wholes. Obviously it does so by controlling individuals, but the level of interaction becomes a crucial one for selection of linguistic structures. Linguistic Communication Is for Coordination Rather Than Understanding A general tendency of the good old-fashioned cognitive sciences to think about cognition as being for understanding the world instead for acting it the world manifests itself also in the cognitive science of language: People communicate through language in order to understand each other. Even in the approaches in which the importance of the dialogical nature of language is recognized, linguistic communication is still viewed as passing information from a speaker to a hearer, in order for it to become common, identical information present in both heads. The goal of communication is to equalize the world models of speakers and hearers (e.g., Pickering & Garrod, 2004). Multiple Systems and Multiple Time-scales of Language Dynamics 45 Alternative view: In the view of language as social coordination (e.g., RączaszekLeonardi & Cowley, 2012) the functional aspect of communication comes to the fore. Paraphrasing Novalis: We will never understand each other but we can do much more than that. People communicate in order to do things together, which only sometimes necessitates identity of reference and/or full likeness of mental models of the situation. Starting with Malinowski’s phatic communication, in which these identities do not seem crucial at all, to task-dependent dialogues in which as long as the actors’ moves in the situation are proper and relevant to the task’s goal no one really cares for identity. An important consequence of this coordinative view of the function of language is that often coordination in realizing a task, that is, functional coordination requires that the roles of the participants are not identical but rather divided, complementary with respect to the task. In order for a proper division of labour to happen, it is more beneficial that the participants’ knowledge, skills and perspectives on crucial aspects of the task are not identical (Rączaszek-Leonardi, 2009). Thus what counts, besides common ground, is the pooled ground that makes dyads or groups more efficient than individuals. Studying Linguistic Meaning Requires Studying Dynamics In connection to the first point about mapping, the mechanism for meaning transfer (or equalizing) between individuals in the process of understanding each other depends almost exclusively on the exchanged words or expressions. Thanks to the mapping relation, meaning in the head of A and B can be gradually equalized by words carrying the elements of meaning from the speaker to the hearer. Alternative view: As pointed out above, the redefinition of symbols as constraints, not containers, changes the definition of meaning of linguistic forms. Congruent with earlier works concerning the symbol-grounding problem (Searle, 1980; Harnad, 1990), according to which the syntactic level of language is not sufficient to account for what language is about, here it is clear that the level of linguistic forms will not constitute an independent explanation. As Pattee writes (1987, p. 337): “it is useless to search for meaning in symbols without complementary knowledge of the dynamics being constrained by the symbols.” The prevalent (Sperber & Wilson, 1986) picture of communication as meaning transfer from person A to B via encoding it in sounds by A and decoding by B does not fit the account. Meaning, rather than being transferred, is made anew in communication, by letting symbols constrain the existing dynamics. Thus in this view on language, the crucial part of specifying the meaning of linguistic expressions is left to existing dynamics of cognitive and coordinative processes: “good biological as well as good engineering design makes the maximum use of natural (non-informational) constraints and laws of nature, so that the control information can be kept to a minimum” (Pattee, 1982b, p. 23). In summary, the characterization of language as a formal symbolic system certainly facilitates some aspects of its study, but at the cost of obscuring fruitful ways to study its vital functions. It takes dynamic events out of the picture, giving the entire explanatory burden to mental symbolic representations. Seeing symbols as 46 Joanna Rączaszek-Leonardi constraints, not vessels for meaning, opens new perspectives but also poses new challenges. In such a view of language, the theory of meaning becomes much less clear-cut and elegant than in the traditional view, where it could be described by intension or extension of a concept with a given linguistic label. One needs to know the system (or rather systems) in which the constraints emerged and know the time scales of changes in these systems, in order to understand how linguistic symbols constrain the existing dynamics. Selected Time Scales and Systems in the Study of Language Already a decade ago, researchers realized that processes pertinent to linguistic and psycho-linguistic phenomena observed in the here and now unfold at many different rates (MacWhinney, 2005; Rączaszek-Leonardi, 2003; Smith, Brighton & Kirby, 2003, 2005). The time scales involved span a wide range of values, from milliseconds of neural events; milliseconds and seconds of individual cognitive events and interactions; hours and days of previous social commitments and social relations; weeks, months and years of language development (ontogeny); years and centuries of cultural language evolution (diachrony); and thousands and millions of years of biological evolution. Relevant events, unfolding on these different time scales concern different systems: from subsystems in the individual brain and body, to individuals, to dyads, groups and populations. Which time scales and systems are the focus of attention depends, obviously, on the research questions asked by a subdomain of language sciences; however, one has to remember that separation of any given system and/or any given time scale might not be feasible: Integrated constraints from all of them act on an individual at any given moment. The framework presented above may, as said before, help to put some organization in this picture of forces, based on the view of symbolic forms as selected functional constraints. An important benefit is that letting dynamics into the picture makes this framework naturally link with some branches of newly developing approaches in cognitive sciences: that of embodied and distributed cognition. Research within these approaches already began to uncover some characteristics of the dynamics on several interesting time scales and systems. Here I briefly describe selected time scales and mention relevant research and research directions. The list below does not strive to be exhaustive. As Enfield (2013) rightly says: There is no definitive set of frames, as new ones may emerge once we find a system of causally linked elements whose changes might be relevant for the study of language. Biological Evolution Time Scale On this time scale, transmittable genetic material is selected, which controls (environmentally dependent) dynamical events responsible for phenotype construction. This is the scale on which, in Gibson’s words senses become perceptual Multiple Systems and Multiple Time-scales of Language Dynamics 47 systems, that is, are tuned to the properties of an environment in relation to an agent’s acting in that environment (Gibson, 1966; Gibson, 1979). From the point of view of language study the crucial aspect is that tuning is not only to action but also to co-action with conspecifics in the environment. Perceptionaction systems seem indeed to be tuned to upholding non-specific interaction (a kind of social glue), by evolving mechanisms that bind together elements of social systems. Research on biological preparation for interaction is usually performed within developmental cognitive psychology, where the underpinnings of early learning processes are identified. This biological preparation seems to involve innate biases for focusing on the other, especially the face and eyes (Field, 1982, Farroni, Mansfield, Lai, & Johnson, 2003), innate synchronization (Murray & Thevarthen, 1985; Nadel, Carchon, Kervella, Marcelli, & Reserbat-Plantey, 1999) and imitation (Field, 1982; Meltzoff & Moore, 1977; Nagy, 2006). It is possible that predispositions for certain granulation of distinguishable auditory stimuli and sequence processing skills also belong to the specific tuning for language use. Other research on language that tackles this time scale explores possible consequences for language evolution of the processes on other time scales, such as perceptual and ontogenetic. Due to the methodological constraints these involved mainly computer simulations (Smith, Brighton & Kirby, 2003) but recently also includes experimental semiotics, which study, for example, the influence of the type of medium on the kind of communicative systems developed (Galantucci, Kroos & Rhodes, 2010), or the properties of cognitive structuring on the emergence of regularities (Cornish, 2010). This program links also to the next time scale of the cultural evolution of language. Cultural Evolution Time Scale Events pertinent to language on this scale have been studied for over a century under the term diachrony. In the view advanced here, language is an important element of a cognitive niche (Laland, Odling-Smee, & Feldman, 2001; Clark, 2006). On this time scale language changes under pressures from the interaction (efficiency of) and individual (economy of production/detection, learnability) levels (see, e.g., Lupyan & Dale, 2010), which serve as criteria for structure selection. Language thus has the power of stabilizing new patterns of individual and collective behaviors much faster than biological evolution, which adapts phenotype to environmental changes. Anthropological focus on how language intertwines with social practice provides a valuable window on the mechanisms behind this stabilizing power (see, e.g., Sidnell & Enfield, 2012; Sinha, in press). Ontogenetic Time Scale Flexibility of human cognition depends on perception-action system being further tuned for interaction and specifically for language use (Cowley, 2003; RączaszekLeonardi, Nomikou, & Rohlfing, 2013). On the basis of the biological preparation for non-specific interaction, culturally established modes of coordination are shaped in 48 Joanna Rączaszek-Leonardi multiple episodes. These educate attention and responsiveness to cues from others. Language is learned in such episodes, by tuning the perception-action system of an individual to utterances of others as affordances for action and co-action (Linell, 2009; Worgan, 2010). On-line Scale of Cognition and Interaction This time scale is perhaps the most researched one because it is connected to conscious experience and because changes on this scale can be readily seen in the lab. According to the third way framework presented above, language consists of physical structures immersed in various kinds of dynamics. On this time scale the natural interactive dynamics that forms the fundament for language action is quite extensively studied. This involves dynamics of natural human synchronization and imitation in non-linguistic interactions (Turvey, 1990; Schmidt, Carello, & Turvey, 1990; Schmidt & Richardson, 2008; Riley, Richardson, Shockley, & Ramenzoni, 2011; for research on neural underpinnings see: Bekkering et al., 2009; Rizzolatti & Craighero, 2004; Tognoli, Lagarde, Deguzman, & Kelso, 2007; Dumas, 2011) as well as linguistic ones (Shockley, Santana, & Fowler, 2003; Dale, Kirkham, & Richardson, 2011). Language both uses bodily-established dynamics, as well as creates specific new ones, as we entrain in syllable rate (Wilson & Wilson, 2005), fundamental frequency and specific turn-taking patterns. Concrete linguistic forms (thanks to their cultural-evolutionary, ontogenetic and experiential history) are then able to steer this basic dynamics into taskand situation-relevant directions (Rączaszek-Leonardi & Cowley, 2012; Fusaroli et al., 2012; Fusaroli, Rączaszek-Leonardi, & Tylén, 2014). Social Events Time Scale Several researchers who acknowledge this multisystem and multiscale nature of language point to yet another crucial time scale which may be difficult to discern because its clock does not tick according to an established time-unit but rather to particular moves in a sequence of social events. This is called sequence time by Goffman (1981, see Enfield, in press), social scale (MacWhinney, 2005) or enchronic frame (Enfield, in press). Meaningful social interactions can be spread in time, preserving its structure, being organized around social values (Enfield, 2013). The ability of humans to rely on perception and memory for such sequences was proposed by Donald as an adaptation for acting in distributed cognitive networks, based on a hypothetical neural process (the slow process [Donald, 2007]). If this scale is also acknowledged, it is evident that at any given moment a human being is enacting multiple such structures, that is, that her behavior is constrained, in parallel, by many forces not readily visible here and now (i.e., at the on-line scale of action or interaction). Other important scales have been mentioned in the literature (for a review see Enfield, 2013; for a discussion of alternative systematizations see Steffensen & Pedersen, 2014). The picture is complicated further by the fact that events unfolding on those different time scales are not independent (Rączaszek-Leonardi, 2009, 2010). Multiple Systems and Multiple Time-scales of Language Dynamics 49 For example, it is the effectiveness of coordination and learnability that decide which structures remain in language on the cultural evolution time scale. On the other hand, the dynamics of here and now, which is harnessed by the structures uttered, is already shaped—both on evolutionary and ontogenetic time scales and both on the individual level of acting in the environment as well as on the level of interaction. Thus, at any given moment, human cognitive system is a point that lies on trajectories of many nested systems, changing on multiple time scales. Dealing with Complexity: Theory of Language in the Form of a Computer Simulation One way to deal with such complexity of an explanatory theory is through a computer simulation (Cangelosi & Parisi, 2002). In the case of phenomena, in which causal loops are contained on multiple time scales and dependent on iterative nonlinear processes, this form of theory allows for testing models and studying the role of parameters. It is very important, however, to emphasize the difference between theories that are based on such simulations and theories as computational algorithms, such as those which were advocated as model psychological theories at the beginnings of cognitive sciences by, for example, Newell & Simon (1976). It is the modern computer’s ability to model stochastic, dynamical, nonlinear, interactive, iterative and nested processes that is the key property on which the theory creations rest, and not the ability to compute according to a preestablished sequence of steps in an algorithm. Limits of Predictability The theory in such a form, that is, a simulation of dynamics on various time scales, which is harnessed by emerging and selected symbolic structures, is inevitably limited in its predictability (Rączaszek-Leonardi, 2014). Even if theorists from various subdisciplines studying language could identify all the relevant time scales and systems, and in a common effort could specify the connections among the time scales and systems, the workings of constraints would still be impossible to predict. This is because constraints are imposed on processes that are dynamic, sensitive to initial conditions and small perturbations, and because the constraints are outcomes of an irreversible selection process. Each one of such processes is thus idiosyncratic, and their generalizability into laws and regularities is limited. Obviously, generalizability and regularity description is possible and desirable, but it is important to keep in mind that in such a system they are only approximations of the workings of the underlying dynamics. This means that they are not lawful generalizations—the regularities always can be violated if the dynamics so dictates. Dependency on history, on irreversible selective processes makes these generalizations different in kind from those used in formulating laws of physics. It is symptomatic that one of the key founders of theoretical bases for the computer metaphor of the mind, in his later work on the processes of morphogenesis, wrote: 50 Joanna Rączaszek-Leonardi Most of an organism, most of the time, is developing from one pattern into another, rather than from homogeneity into a pattern. One would like to be able to follow this more general process mathematically also. The difficulties are, however, such that one cannot hope to have any very embracing theory of such processes, beyond the statement of the equations. It might be possible, however, to treat few particular cases in detail with the aid of a digital computer. This method has the advantage that it is not so necessary to make simplifying assumptions as it is when doing a more theoretical type of analysis. ... The essential disadvantage of the method is that one only gets results for particular cases. (Alan Turing, 1952, pp. 71-72; emphasis mine) This dependence on “results for particular cases” which, at best, give fuzzy classes of solutions—some of the time—show where the limits of predictability are. Models of language, as a system of replicable constraints, suffer from—or are blessed with—the same limitations. Perhaps the claim about the irreducibility of historical trajectories of language processes to deterministic paths is rather obvious. However reaching this claim from the direction motivated by discovering the significance of dynamics in symbolic processes, gives a place for symbols in these dynamics and perhaps leads to a better understanding the sources of this unpredictability on one hand and the necessity for it on the other.