Experimental Design And Experimental Inference In Stone .
J Archaeol Method TheoryDOI 10.1007/s10816-017-9351-1Experimental Design and Experimental Inferencein Stone Artifact ArchaeologySam C. Lin 1 & Zeljko Rezek 2 & Harold L. Dibble 2,3,4# Springer Science Business Media, LLC 2017Abstract Lithic researchers rely heavily on experimentation to infer past behaviors andactivities based on stone artifacts. This paper explores the analogical nature of archaeological inference and the relationship between experimental design and inferencevalidity in stone artifact experimentation. We show that actualistic flintknapping lacksvital aspects of scientific experimentation, and thus has inherent inferential issues ofanalogical adequacy and confidence. It is argued that a greater emphasis on hypothesisconstruction and variable control is needed in order to establish sound referentiallinkages upon which constructive analogic inferences about the past can be built.Keywords Experimental archaeology . Experimental inference . Stone artifacts .Analogy . Inference validityIntroductionExperiments have long played a central role in the development of lithic studies. In thelate nineteenth century, experimental replication of prehistoric stone artifacts was usedto demonstrate their anthropogenic origin, and, by extension, the antiquity of humankind (Johnson 1978). During the 1950s and 1960s, the work of Crabtree, Bordes,* Sam C. Linsamlin@uow.edu.au1Centre for Archaeological Science, School of Earth and Environmental Sciences, University ofWollongong, Northfield Avenue, Wollongong, NSW 2500, Australia2Department of Human Evolution, Max Planck Institute for Evolutionary Anthropology, DeutscherPlatz 6, 04103 Leipzig, Germany3Department of Anthropology, University of Pennsylvania, 3460 South Street, Philadelphia,PA 19104, USA4Institute of Human Origins, Arizona State University, School of Human Evolution and SocialChange, P.O. Box 874101, Tempe, AZ 85287, USA
Lin et al.Tixier, and others further brought lithic experiments to the forefront of lithic studies. Inthe following decades, the field saw a surge of experimental studies exploring variousaspects of lithic technology through the replication of stone artifact forms, from basicproperties of fracture mechanics and flake formation (e.g., Cotterell and Kamminga1987; Cotterell et al. 1985; Dibble and Whittaker 1981) to lithic variability related topercussion techniques (e.g., Barham 1987; Flenniken 1987; Kobayashi 1975;Newcomer 1975; Sollberger 1985; Speth 1974), reduction strategies (e.g., Amicket al. 1988; Flenniken 1978; Magne 1985; Newcomer 1971; Sollberger and Patterson1976), and resharpening (e.g., Flenniken and Raymond 1986). Research areas alsoexpanded from the production of stone artifacts to topics of function, use, and efficiency(e.g., Crabtree and Davis 1968; Fischer et al. 1984; Hayden 1979a; Kamminga 1980;Keeley 1974; Odell and Odell-Vereecken 1980; Sheets and Muto 1972; Walker 1978).Today, lithic experiments come in a variety of forms involving different methods andresearch designs to address a wide range of questions. Increasingly, studies also useexperimental reconstructions to address issues beyond the immediate production anduse of artifacts, including the identification of technical systems, end-product designand production (e.g., Boëda 1993, 1994, 1995; Boëda et al. 1990; Delagnes andMeignen 2006; Meignen et al. 2009; Mourre et al. 2010; Scimelmitz et al. 2011), skilland knowledge transmission (e.g., Eren et al. 2011; Geribàs et al. 2010; Nonaka et al.2010), and the potential selective pressure on hominin cognition and biomechanicsassociated with the habitual production and use of stone tools (e.g., Key and Lycett2011; Key 2016; Stout et al. 2000, 2014; Williams et al. 2012).Researchers increasingly rely on experiments to draw inferences about homininbehavior and evolution from stone artifacts. However, much of the discussion about thetheoretical underpinnings of such experiments, and their appropriateness for generatingvarious levels of archaeological inference remains limited. Specifically, the confidenceand legitimacy of the experimental inference is strongly contingent on the design of theexperiment (Eren et al. 2016) as well as the underlying analogic premise (DomínguezRodrigo 2008). This sort of inquiry into the nature of archaeological reasoning is notnew, and by now it has a relatively long history starting by large in the 1960s and 1970swith a number of archaeologists shifting their focus to the philosophy of science (e.g.,Fritz and Plog 1970; Watson et al. 1971). This shift was largely driven by the desire toestablish firm referential frameworks, consisting of Bmiddle-range theories, to connectthe static archaeological record to the dynamic yet unobservable past (Binford 1962,1977, 1981; also see Raab and Goodyear 1984; Schiffer 1998). In these discussions, therole of experimentation was emphasized as part of the Bhypothetico-deductive processof testing and falsifying existing assumptions of archaeological interpretation (Ascher1961b; Schiffer 1975). When a hypothesis resists falsification under experimentaltesting, it is viewed as potentially valid in the sense that the underlying principle cancontinue to be used for drawing archaeological inference until falsified by furthertesting (Outram 2008). Validity can be defined as Bthe best available approximationto the truth or falsity of propositions (Cook and Campbell 1979:37).In his original publications of Archaeology by Experiment (Cole 1973) and Experimental Archaeology (Cole 1979), Cole devised a series of rules for the design ofarchaeological experiments to ensure a general level of inferential rigor and reliability.In terms of lithic experimentation, over the years, studies have underscored theimportance of experimental design, variable control, and the nature and adequacy of
Experimental Design and Experimental Inference in Stone Artifact.experimental inferences (e.g., Amick et al. 1989; Carr and Bradbury 2010; Lycett andEren 2013). In a recent review, Eren et al. (2016) argued that replicative lithicexperimentation is an important approach within a broader hypothesis-driven archaeology framework for testing hypotheses, establishing models, and validating analyticalmethods.We wholeheartedly agree with the emphasis on the importance of experimentaldesign in lithic experimentation. However, we believe the theoretical connectionbetween experimental design and the broader framework of archaeological inferencegeneration has been largely overlooked in these discussions. This ambiguity has directimplications on not only the validity of experimental inference, but also the applicability and comparability of the resulting interpretations of past behavior. For instance, ina discussion of zooarchaeological experiments, Domínguez-Rodrigo (2008) demonstrated that variation among experimental outcomes can be attributed to differences ofassumptions in experimental design and the underlying analogic premise rather than thestudied material per se. This observation illustrates that, despite having a hypothesisdriven design, the validity of experimental outcome and interpretation is still heavilydependent on (a) assumptions of archaeological inference, (b) the nature of theformulated hypotheses, and (c) how the experiments are designed to test the hypotheses. It is thus critical to disentangle the theoretical interaction between experimentaldesign and archaeological reasoning.Focusing on lithic studies, this paper examines the properties of experimental designand their interaction with the creation of experimental inference related with stoneartifact archaeology. In order to do this, it is first necessary to consider the nature ofarchaeological reasoning in the form of analogy. We then compare flintknapping as anactualistic approach to scientific experimentation with respect to the issue of confounded variables and the ways that uniformitarian assumptions are treated in the formulationof hypotheses. This analysis is accompanied by discussions regarding the basic properties of experimentation, including variable control, sources of error, and inferencevalidity. Finally, we differentiate between Bpilot versus Bsecond generation experiments and discuss their respective roles in the process of archaeological knowledgegeneration.Analogy in Archaeological ReasoningFor archaeology, the processes that led to the formation of the material outcomeobservable today (the archaeological record) cannot be directly experienced. Unlessthere are other sources of information, such as historical documentations, we rely onanalogy to make inferences about the past by linking concepts and relationships derivedfrom the present to aspects of the archaeological record (Binford 1981; Clarke 1968;Gifford-Gonzalez 1989; Wylie 1985). In her seminal paper, Gifford-Gonzalez (1991,also see Wylie 1985), differentiated between the use of Bformal analogy and Brelationalanalogy in archaeology. Based on the ordinary experience of the observer, formalanalogy operates by drawing a causal connection between observed modern processand its material outcome. When archaeological items share formal similarities with amodern object, it is assumed that the observed process in the contemporary world alsooccurred in the past. The operation of formal analogy is summarized schematically by
Lin et al.Gifford-Gonzalez (Fig. 1) and contains three key assumptions: (1) the linkage betweenthe observed modern object and process is causal, (2) the formal similarities between themodern and prehistoric objects are meaningful to the analogical inquiry, and (3) theinferred process is uniformitarian and thus can be projected to the past.Most archaeological systematics is rooted in formal analogy. Because the archaeological record is largely an anthropological phenomenon, it is intuitive for archaeologists to use their daily experience of the human world as primary analogs for makingsense of archaeological remains (Holdaway and Wandsnider 2006; Plog 1974). Thisform of analogy can be seen in the basic naming and description of artifact categories(e.g., Bscraper, Bknife ) (Clark 1989) to interpretations of past events and processes.The most explicit demonstration of formal analogy is the early approach of ethnographic analogy, where living groups are chosen as counterparts of past societies basedon formal similarities in either material culture or ecological and subsistence conditions(Ascher 1961a; Stiles 1977).In contrast with formal analogy, relational analogy appeals to uniformitarian relationships to justify the inferred analogic link. While formal analogy does involve arelational component in linking modern analogs to the subject in question, the relationship is singular, causal, and based largely on observed formal similarities (e.g., aflake can be made by hitting a stone with a hard material. Therefore, an artifact sharingsimilar attributes to a flake was, by extension, produced from a stone being hit by ahard material at some point in the past). On the other hand, relational analogy isobtained through referencing other known knowledge that is Bsystematic and causallybased (Gifford-Gonzalez 1991: 224). Therefore, the identification of an archaeologicalflake is not simply because the item possesses qualities that have been deemed relevantby the modern observer, but rather the item contains qualities that indicate knownprocesses, such as fracture mechanics and solid geometry, that we can expect to operateuniformly across space and time. In effect, the condition on which uniformity isassumed has been made explicit in relational analogy. By invoking a particularrelational condition for constructing an analogic inference, we are stating that thisassumed premise enables the observed formal similarities to testify similarities in otheraspects of past phenomena (Wylie 1988).Fig. 1 Gifford-Gonzalez’s model of analogical reasoning in formal analogy. Blank boxes indicate observations made in the present world. Redrawn after Gifford-Gonzalez (1991:222)
Experimental Design and Experimental Inference in Stone Artifact.This relationship between uniformitarianism and analogy was developed into ahierarchical order of inference by Hawkes (1954), who argued that uniformity can beexpected in facets of human life that are restrained by the physical world, such assubsistence, economy, and technology. Thus, as one moves up the inferential order tothose of the socio-political and ideological realm (i.e., intentionality, social norm,cultural tradition), the reliability of the analogic inference drop significantly due totheir greater reliance on culturally specific manifestations. In the 1960s, Binford (1962,1977, 1981) conceptualized these relational conditions or linkages as Bmiddle-rangetheories and they are required to have the following attributes (Wandsnider 2004).First, the causal connection between the material phenomenon and the generatingprocess has to be unambiguously demonstrated and documented. Second, this causalconnection has to be uniformitarian in nature so the inference can be projected into thepast with warrantable confidence. The third attribute is that middle-range theory has tooperate independently of ideas about the past that archaeologists wish to investigate,and thus can serve as a neutral medium for inferring the occurrence of past processesfrom the archaeological record.The point of discussion here is not to equate relational analogy with the Binfordiannotion of middle-range theory. Instead, the goal is to emphasize the importance of theassumed premise of uniformity that bridges the analogic inference. Few would disagreethat archaeological materials operate on physical principles and these principles areuniformitarian in nature (Eren et al. 2016). Therefore, it makes sense that archaeological reasoning can rely on these physical processes as means to draw inferences aboutthe formational mechanism through relational analogy. However, an important obstaclein the validity of relational analogy is equifinality, where the same material outcomeand physical process can be initiated by different agents under dissimilar context. Takethe formation of flake as an example, flake attributes such as platform and bulb ofpercussion allow us to infer that the object was produced by conchoidal fracture due toforce impact. Yet, under what manner (e.g., percussion technique, manual gesture,reduction sequence, cognition) was the flake produced may be beyond the range ofinference justifiable by the underlying relational premise. To overcome this analogicissue, researchers have proposed the use of additional lines of evidence to increase theconfidence of archaeological inference and to reveal ambiguities in explanatory models(Wylie 1985, 1989, Binford 1967, 1981, 1987; Gifford-Gonzalez 1991, Bradbury andCarr 1995, 2004). By connecting multiple analogic linkages at different analyticalscales, it is possible to develop internally coherent bodies of referential knowledge(Wandsnider 2004) and move archaeological interpretation beyond the constraint of ourmodern-day knowledge (Binford 1981; Wylie 1985, 1988; Gifford-Gonzalez 1991;Bardbury and Carr 1995, 2004).Experimentation in Lithic StudiesUnder this framework of relational analogy, the strength and validity of inference isgoverned by the assumed uniformity of the premise that underlies the analogic inference. In the 1970s and 80s, experimentation was advocated as one of the principleapproaches for establishing relational linkages in archaeological interpretation. However, contrasting with the conventional definition of experimentation, these
Lin et al.archaeological studies align more with actualistic research of discovering and evaluating archaeological relevant variables within living contexts. In fact, in one of theoriginal articulations of middle-range theory, Binford (1981) urged archaeologists toconduct actualistic studies for generating possible referential knowledge. The source ofthis emphasis on actualistic research can be attributed to the insistence on the analogicalnature of archaeological reasoning. Because past processes can only be comprehendedthrough modern observations, it makes sense that a general understanding of theseprocesses can be gained through replicating past activities and behaviors.This sentiment is particularly apparent in stone artifact research. Since few societiestoday are dependent on the use of stone tools on a daily basis, lithic technology andstone artifacts in general remain largely a body of knowledge that is essentially alien tomodern archaeologists. While ethnographic accounts provide valuable informationregarding the ways stone tools were used in modern living contexts, their directapplicability to archaeological materials remains limited as they often represent a subsetof technological behavior operating within a specific cultural context and time scale. Assuch, stone artifact researchers have come to rely on replicative experiment as aprinciple means to understand the behavioral processes that underlie the formation ofstone artifacts. Prior to the twentieth century, scholars used experiments primarily tosupport the anthropogenic origin of ancient chipped stone implements. This was donein part due to the recognition that flakes can be produced under natural conditions andthus the need to discriminate cultural artifacts from naturally flaked stones (Evens1872; also see Johnson 1978; Lerner 2013;). In the late nineteenth century, Holmes(1894) employed experiments in his study of American bifaces as a way to gain anunderstanding of the production sequence of artifacts from raw material acquisition,through technical production, to ultimately arrive at forms recognized in the archaeological record. This notion of sequential manufacture to produce a particular endproduct became a key element of modern lithic experimentation (Bleed 2001).In the first half of the twentieth century, increased attention was given toflintknapping and the documentation of various knapping techniques for replicatingartifact types that are comparable to those observed archaeologically (Johnson 1978;Lerner 2013). This approach to lithic experimentation was further popularized andincorporated into the realm of mainstream archaeological investigation in the 1960s bythe work of several skilled flintknappers, including Crabtree, Bordes, Tixier, Callahan,and Bradley. Increased consideration was also paid to the reporting and/or control ofvariables involved in knapping experiments, such as the size and nature of the rawmaterial, the technique of production, and the properties of hammerstones used(Johnson 1978). However, much of the literature on lithic replication during this timefocused on identifying the how-to or craft aspects of replicative flintknapping ratherthan answering specific archaeological questions (Andrefsky 2005).Since the 1980s, the focus of experiments shifted from end-product manufacture tothe production of by-products as well as the overall reduction sequence. Specifically,studies examined the interrelationship between different knapping sequences with thecharacteristics of the overall produced lithic assemblage. This is best represented by thereduction sequence approach of North American archaeologists (Bleed 1996; Dibble1984, 1987, 1995b; Frison and Raymond 1980; Magne 1985; Morrow 1997) and thechaîne opératorie school of France and Continental Europe (e.g., Boëda 1986, 1988;Boëda et al. 1990; Geneste 1985; Pigeot 1990; Sellet 1993). These studies stress the
Experimental Design and Experimental Inference in Stone Artifact.importance of the techno-behavior process that underlies the formation of lithic artifacts, which could effectively be captured through refitting and replicativeflintknapping experiments. One of the main objectives of these experimental projectswas to transcend typological units and to investigate holistically the procedur
activities based on stone artifacts. This paper explores the analogical nature of archae- ological inference and the relationship between experimental design and inference validity in stone .
2.3 Inference The goal of inference is to marginalize the inducing outputs fu lgL l 1 and layer outputs ff lg L l 1 and approximate the marginal likelihood p(y). This section discusses prior works regarding inference. Doubly Stochastic Variation Inference DSVI is
Stochastic Variational Inference. We develop a scal-able inference method for our model based on stochas-tic variational inference (SVI) (Hoffman et al., 2013), which combines variational inference with stochastic gra-dient estimation. Two key ingredients of our infer
Experimental and quasi - experimental desi gns for generalized causal inference: Wadsworth Cengage learning. Chapter 1 & 14 Campbell, D. T., & Stanley, J. C. (1966). Experimental and quasi -experimental designs for research. Boston: Hougton mifflin Company. Chapter 5. 3 & 4 Experimental Design Basics READINGS
stochastic inference, not deterministic calculation AI systems, models of cognition, perception and action Parallel Stochastic Finite State Machines Probabilistic Hardware Commodity Hardware Specialized Inference Modules Universal Inference Machines Mansinghka 2009 Universal Stochasti
Statistical Inference: Use of a subset of a population (the sample) to draw conclusions about the entire population. The validity of inference is related to the way the data are obtained, and to the stationarity of the process producing the data. For valid inference the units on which observations are made must be obtained using a probability .
\Learn to use the inference you will be using" (usually with variational inference). 3 Just model each p(y c jz) (treatlabels as independent given representation). Assume that structure is already captured in neural network goo (no inference). Current trend:less dependence on inference and more on learning representation.
2 Classical mean-field variational inference 3 Stochastic variational inference 4 Extensions and open issues (Hoffman et al., 2013) . Stochastic variational inference 4096 systems health communication service billion language care road 8192 service systems health com
asset management must be considered as one of the first revolutions in financial technology. However, it quickly became the industrial secret of many successful hedge funds such as Re-naissance, D.E.Shaw, Two Sigmas, CFM, e.t.c. The 2008 crisis has changed the investment point of view of investors and the regulators. They required more and more efforts from the hedge fund industry and asset .
