On the Epistemic Potential of Virtual Realities for the Historical Sciences. A Methodological Framework

Virtual or augmented audio-visual environments can be employed not only for the impartment of knowledge to a wider audience, but also for the generation of knowledge within the historical sciences. In this context, the transformation of numerical models of historical circumstances into an immediate sensual experience may be used both in an exploratory manner as well as for testing specific hypotheses through subjective perceptual analysis. As with any other empirical approach, the new insights provided can be biased on different levels. In order to make virtual realities (VRs) a valuable tool for research, traditional quality criteria for empirical research need to be adapted to the specific setting created by observational fieldwork ‘within’ 3D audio-visual computer simulations. Two major causes for degradation in the credibility of VR-based historical research are related to the relationship between simulation and historical ‘reality’ as well as to the human agents experiencing the simulated environments and the conclusions drawn from their subjective impressions. Hence, our contribution attempts to outline procedures and methods for estimating and comparing the ecological validity of virtual environments as well as the level of intersubjectivity regarding the inferences drawn during and after experiencing them. For this purpose, we suggest to synthesize existing ideas and procedures originating from virtual reality research, media psychology, communication science and ethnology.

18 th -century Venice (Lercari 2016), or of audio-visual media installations such as the Poèmeélectronique and the Philips pavilion by Le Corbusierand EdgardV arèse at the Brussels world fair in 1958 (Lombardo et al.2 009).
Less frequent are attempts to generate new knowledge from the virtual reconstruction of historical scenarios,a nd much greater are the reservations within the scientific community against this kind of scientific practice. These reservations maybedifferentlymotivated in each individual case. In their core, however,they are related a) to doubts regardingthe credibility of virtual environments as af orm of scientifice vidence; b) to the question of how new knowledge may emerge from virtual environments,which are always basedo na lreadye xisting historical evidence;a nd c) what kind of role virtual or augmented realities could playwithin an epistemological concept of historical research. With the current text,wew ould like to sketch out am ethodological framework thatw ill attempt to provide answers to these questions. This framework will be exemplified by ac oncrete problem drawnf rom the authors' own research. The open question concerned a) the maximum audience size reached by as peaker on the ancient Forum Romanumb etween ca. 500 BC and ca. 500 AD; b) how the size of this audience changed with the several modifications and relocations of the Rostra; and c) whether these modifications were carried out for reasons of acoustics,orrather for the political and symbolicmotivesu sually invoked in the historiographyo ft he Roman empire. This concrete problem was part of al argerp roject dedicated to the digital reconstruction of the ancientF orum Romanum¹ and worked on by at eam of archeologists and acousticians.
In the following outline for am ethodologicala pproach towards the use of virtual realities for historical research, we will restrict ourselvest ot he acoustic domain, which was particularlyi mportant for the givenp roblem, although all technical and methodological stages have an almosto ne-to-one equivalent in the visual domain. Since manyofthe methodological problems are closelyrelated to the technical process of generatingavirtual environment,wewill first give abrief outline of binaural synthesis as the chosen approach for virtual acoustic reality (Fig. 2).

Virtuala coustic environments
Binaural synthesis is one approach for the generation of virtual acoustic environments. It is the acoustic equivalent of stereoscopic displays,inwhich one image is generated for each eyeofasingle observer. In binaural synthesis, one acoustic signal is generated for each ear of as ingle listener and presented by headphones. These signals are designed to excite the eardrum as an acoustic sensor in the samew ay as ac orresponding real sound field.
Fort he synthesis of this signal, a3 Dm odel of the desired environment is used together with as pecification of the acoustic properties of the desired sound sourcea nd the desired receiveri no rder to numerically simulate the sound propagation from sourcet oreceiver. The resultofthis simulation, including both the direct soundaswell as all sound reflections at the boundaries of an open or enclosed space, is encoded as binaural room impulseresponses (BRIRs), which describet he transfer path between as ound sourcea nd ad efined (e. g.: human) receiver. If these impulseresponses are linked to at emporal sourcesignal such as speech or music by an umerical process called convolution,the correspondingsound sourcewill be perceivedatthe desired point in space encoded in the BRIR.T his perception includes all spatial properties of both the source and the environment,s uch as the distance, the direction of the sourcea nd the size and reverberation of the surrounding space. In order to invoke this perception, the resultofthe convolution has to be applied to the ear canalofalistener by headphones. The generated auralization (Vorländer 2008) of an acoustic scene will onlyb ep erceiveda si mmersive if the listener can naturallyi nteract On the Epistemic Potential of VirtualR ealities fort he Historical Sciences with the produced sound field, i. e. if the ear signals are re-calculated for different head orientations, so that the listener can move within the scene rather than the scene moving with the listener.This feature alsohas an equivalent in stereoscopic displays,inwhich the imagehas to be re-calculatedinreal-time whenever the observer changes his or her visual orientation. And just as with stereoscopic displays,i td epends on the reproduction device, whether the simulation is presented as virtual or Augmented Reality.H eadphones can exhibit different degrees of openness,r angingf rom closed headphonesw ith as trongi nsulation against the external sound field, occasionallye vens upported by noise-cancelling algorithms, to open or even extra-aural headphones, which leave ag ap of some centimeters between the ear and the headphone and constitutenosignificant obstacle for the external sound field.The latter are particularlysuitable for simulations presented as Augmented Reality (see fig.3 ).   If the observations made within such avirtual acoustic reality shall be exploited as eyewitness reports about the correspondingh istoricals cenery,the credibility of resulting inferences can be questioned at two levels. The first concerns the relationship between the simulation and the historical 'reality',t he second concerns the reliability of the observations made by human observers inside the generated virtual or Augmented Reality.B oth aspects will be considered more closelyi nt he following.

Considering the uncertainty of virtual reconstructions
If virtual reconstructions are evaluated as aresearch tool, it is necessary to specify their epistemologicalfunction, since -as with anyother tool -their adequacy can onlyb ed etermined with respect to ad esired goal. In the caseo fh istorical research, this requires aw ell-specified research question or hypothesis that can be answered of confirmed on the basis of qualitative or quantitative data retrieved from virtual historicale nvironments. In the example discussed throughout this text,t he question concerns the speech intelligibility in the ancient Forum Romanum,i no ther cases it might relate to the soundi mpression of historical concert halls, the visual impression of historicalc hurches,the visual impression of paintingsi nm useums before and after the introduction of electric light or others imilar problems.
In the present discourse about the possibility of research in virtual environments, we think that the discussion is often overlyf ocused on the question of whether simulations can replacet he experience of ar eal, historical environment,o r, more generally, whether virtual reality can fullyr eplace 'real reality'. This question seems unproductive to the authors. This is less because it might be technicallyu nfeasible in the foreseeable futuret op rovide fullyt ransparent virtual environments.³ Morer elevant from am ethodological point of view is that there is no criterion to answer this question without specifying the task, that an observerof(and listener 'in')the virtual or Augmented Reality has to perform. Form ost questions of mere perceptibility (e. g., speech intelligibility such in the present example), however,s uch criteria can be well-formulated and justified -based on theory or on empirical pretests -for amedia system in order to fullyc onveyt he necessary cues for al istener confronted with this task.
Consequently, it might not always be necessary to generate virtual environments with amaximumlevel of realism and interactivity (whatever thatimplies). Simplyw ith respect to efficiency and limited resources, it might be rather desirable to provide avirtual environment with onlythe featuresnecessary to answer the question for which it was devised.

Reliability and biasI :T he virtuale nvironment
As far as the process of transformingh istorical evidence into the virtual reality presentation of acertain historical scenario is concerned, each stepofthe transformationcan introduce uncertainties which affect the credibilityofthe final result.I fw et ake ac loser look at each of these steps in the order they have to be processed in practice ( fig.2), the first stageincludes the establishment of an en- The "perceptual illusion of non-mediation" was suggestedasacriterion for completelytransparent virtual realities providinga nu nrestricteds ense of presence ( Lombard/Ditton 1997, 9). vironmental model of the desired scenario. In visual simulations this would include a3 Dm odel incorporating the geometry of the space and the texturingo f the surfaces.Inacoustic simulations it would includea3D model and the acoustic properties of all boundaries such as the absorption and scattering coefficients of the employed surfaces.The types of model parameters required at this stage depend on the subsequent simulation algorithm; while for wave-based numerical simulations of the sound propagation such as the finite element method (FEM) acomplex impedance is required for each boundary element representing the amplitude and phase differenceb etween the incoming and the outgoing sound wave,f or ray-based acoustical simulations, oftenc ombiningr aytracing and image-source-method algorithms, onlyareal-valued reflection coefficient is required, as the phase of the acoustic wave is ignored in this simulation type.
No matter which simulation approach is selected, anyoft he required input parameters is subjectt ou ncertainty which mayr esult from insufficient knowledge of the historicalc onditions themselveso rf rom insufficient knowledge of how to translate historicale vidence into the required input parameters.I nt he case of the Forum Romanum, for example, there wasi nsufficient knowledge about the material used for the walls of the surrounding buildings. Furthermore, also the absorption and scatteringp roperties of Romanc oncrete (opus caementicium), which was introduced after 200BCand usedfor manyofthe largerbuildingsofthe forum, could onlybeestimated. The exact state of construction of the forum at as pecific point in time was partlyu nknown, for there were contradictory information in different sources such as pictorial representations and textual sources. Further,w hile it was known that certain sounds ources (speakers, horse-drawnv ehicles, other sources of noise) werep resent in the investigated scenario, their exacta coustic radiation characteristics, i. e. the frequency-dependent directivity,m ight be unknown.
The uncertainties at this stage, which we suggest calling modeling uncertainty,r esultfrom the classicalproblem of incompleteori nconsistent historical evidence. They are not yetrelated to the computational processing of this evidence, but have to be understood and treated with aclassical source-critical approach; the simulation method onlydefines the type of evidence required in order to generate avirtual representation. Historians might otherwise not be interested in the sound absorptiono fR omanc oncrete; for the acoustic simulationi ti se ssential information.
At the second stage, the reliability of the numerical simulation itself maybe questioned, i. e. the assumption that the sound propagation from the sourcet o the receiveri sc orrectlym odeled by the selectedn umerical approach. We could term the uncertainties at this stage simulation uncertainty. Simulation uncertainty is an engineering problem inherent to anykind of numerical simulation of physical processes. It is usually addressed by using measurements conducted on real physical systems as ar eference and by comparing these to simulations on the basis of specific dependentvariables. These variables are selected according to ad efined functional application. Fore xample, if it is the functiono fa n acoustic simulation to predict the reverberation of newlyd esigned concert halls, measured reverberation times would constitutea na ppropriate reference for simulated reverberation times, as would otherr oom acoustic parameters (Bork 2005a(Bork , 2005b. In the case of virtual realities intended to createaconvincing sensoryimpression, the perceiveddifferencetoacorrespondingreal environment would be an appropriate reference for an evaluation of the simulation uncertainty.F or the operationalization of the 'perceivedd ifference',d ifferent measurable indicators have been proposed (see next section).
At the next stage, the results of an umerical simulation have to be encoded in acertain data format for storageorf or live transmission, and decoded for reproduction. As with anycommunication channel, the information transmittedby encodingand decodingisoverlaid by noise, which can have amultitude of sources.Inthe case of binaural synthesis, these include, for example, the spatial discretization of the sound field at the receiver. Whereas in ar eal sound field, listeners can introduce infinitesimallys mall modulations to the sound field by infinitesimallysmall head movements, BRIRs are available onlyfor apredefined grid of head orientations. They have to be interpolated for heado rientations in between. The simplest solution is an earest-neighbor interpolation, i. e. the hard switching in the predefined grid; but even more advanced solutions are always error-prone approximations. Other encoding/decodinge rrors can be duet ot he fact that onlyi nitial parts of the impulse responses are exchanged dynamically to increase computational efficiency.They can also result from the audio signal format usedo rs imply from the numerical resolution of the digital system used. The errors introducedbyencodingand decodingcould be summarized as coding uncertainty.
At af inal stage, the accuracyoft he optical or acoustic reproduction can be questioned, i. e. the extent to which the qualityoft he optical or acoustic signal presented to the eyes or the ears with respect to the desired referenceisdegraded by the employed reproduction system. Whereas codinguncertaintyisintroduced in the digital domain, for reproduction some kind of human-computer interface is required. In the case of binaural synthesis, degradations can relatet ot he headphonesa nd the headt racking device used. The transfer function of the headphones, for example, can lead to timbrald ifferences between the original and the technicallyreproduced sound source. Latencies in the acoustic adaption of the ear signals can lead to spatial instability of the reproduced sound source, latencies in the visual adaption in head mounted displays can even cause so-called simulator sickness if visual and vestibular motion cues are no longer consistent.I naccuracies at this stagec ould be called reproductionu ncertainty.

Characterizing the reliability of virtual environments
The uncertainties introduced at the modeling,simulation, coding and reproduction stageh avetobei ndicated and quantified by appropriate means in order to conveyatransparent overall picture of the reliability of the virtualization. At the modeling stage, incomplete knowledge about the original spatial environment can, for example, be documented by alternative models representing different potential historicalc onditions,a ll of which can be plausible with respect to the archeological remains and historical knowledge in general. Formodeling parameters such as the acoustic boundary conditions,a ne stimated range of possiblevalues can be givenaccordingt oexpert knowledge.There are different ways to investigate the propagation of these uncertainties at the level of input parameters through the simulation to output parameters such as,i no ur example, speech intelligibility in the Forum Romanum. Ford iscretei nput parameters,such as different models representing different possible spatial configurations, as imulation can be run for every element of as et of models, giving a corresponding set of output parameters.For input parameters with acontinuous estimatedrangeofuncertainty, asampling scheme maybeapplied, and then the simulationh as to be run for this sample of input parameterv alues in order to estimate the outputu ncertainty.F inally, as tatisticallyc orroborated estimation of the uncertainties at the level of the output parameter would specify input parameters as probability distributions and use as ampling scheme that considers the probability of each parameterv alue. Thisc an be achieved by random sampling (Monte Carlo strategies) or by dividing the rangeo fe ach uncertain input parameter into equi-probable intervals and sampling accordingly (Rubinstein/ Kroese 2008). The resultisthen aset of output values representing the probability distribution of the parametero fi nterest (cf. fig.4).
The uncertainties arising from the numerical simulation algorithmitself are usually estimated by comparingsimulations to measurements of the corresponding real physical systems. This is not as straightforward as it mays eem, however,b ecause the measurements used as references for the uncertainty of the simulationhaveameasurement uncertainty themselves. Moreover,itcan be difficult to ensure that the input parameters used for the simulation will correspond exactlyt ot he parameters of the real system. And, finally,b enchmark tests be-tween measurement and simulationa re, at least in acoustics,b ased on considerable technical efforts. Hence, they can be conducted onlyfor asmall sample of physical systems and provide onlyarough estimation of the reliability of the simulationalgorithm without statistical information about the uncertainty distribution. Nevertheless, these benchmarksare common for most numerical simulation methodsi nt he formo fr ound-robin tests or competitions between alternative approachesa nd implementations.⁴ Encodingand decodingerrors can be controlled by system design in order to keep them below apredefined technical or perceptual threshold. The spatial discretization of binaural simulations, for example, has been shown to be inaudible with asampling grid of 2°for the horizontal, vertical and lateraldiscretization of BRIRs (Lindau/Weinzierl 2009). With respect to the dynamic exchangeo ft hese impulse responses for different head orientations, ap rediction formula has been devised specifying the time window necessary for such ad ynamic exchangei no rder for audible artifacts to be avoided . The underlying test procedures are similar to those used for the encodingand decoding of audio signals with data compression (mp3), which determine at hreshold of transparency, i. e. the data rate necessary for the differences between the encoded/decoded sourcem aterial and ag iven referencet ob ei naudible or come with acceptable degradations.
At the reproduction stage, it is the established parameters for imagingdevices or sound transducers that can be applied to evaluatethe information loss between the computer model and the human user. These include the field of view or the imager esolution of displays or the frequency bandwidth, the linearity of the transferf unction or non-linear distortions causedb yheadphones. Fori nteractive virtual environments, it includes the update rate or the system latency of the head tracking device. Fora ll these parameters therei salarge stock of research in the field of quality and usability research that can be used in order to assess the perceptual relevance of the intrinsic limitations of the systems and devices used.
Forv irtual acoustic environments as aw hole, different degrees of distinctness of simulation and reality have been suggested as measures:The plausibility of as imulation , measuring the ability of as ubject to identify the simulation in ar andom, alternating presentation of simulation and reality,a nd the authenticity of as imulation , measuring the ability of as ubjectt op erceive any differenceb etween simulation  Fore xampless ee http://rr.auralisation.net/ for room acoustic simulation and auralization, and Durante/Riedel (2008) for wind flow simulation. and reality,evenifthe simulationcannot be identifiedassuch. Foradifferential diagnosis of differences between simulation and real references, ataxonomyhas been developedbyqualitative research . While these tests give no indication of the stageatwhich the virtualization degradations occurred, they give an overall picture of the quality provided. Fig. 4: Uncertainty propagation through the acoustic simulation. Top: Assumingt hatt he absorption coefficient of the audiencel ies in an interval between 0.5 and 1.0 (with at riangular distribution, and 0.75a st he most probable value, topl eft), the resulting uncertainty of the sound power levela tt he bordero fi ntelligibility is ca. ±1 dB (5 %/95 %q uantiles, indicated as dashed lines, topright). Bottom: The color mapshows the regions of good (red) and poor (blue) speech intelligibility on the ancient Forum Romanum.T he solid line indicates the border of intelligibility,w itht he dashed lines related to the uncertainty of ±1 dB soundp ower level calculated above.T he corresponding sizeo ft he audiencea reat hatc ould be reached by a trained speaker is 3400 m 2 (2800 m 2 /3800 m 2 ).
On the Epistemic Potential of Virtual Realities fort he HistoricalS ciences

Reconstructing historical eventsv s. reconstructingh istorical spaces of possibility
The abovem entioned uncertainties at the different stages of digital reconstructions will, as aresult,lead to acorresponding rangeofuncertainty for the specified variables of interest,s uch as speech intelligibility in the described model study. Independent of this problem is the question of whether the generatedvirtual environment and the results derivedf rom it are supposedt or epresent one specific historical scenario, or whether they aim at a range of possibilities related to ac ertain historicals ituation. Are we interested in the size of the audience able to understand Marcus Tullius Cicerod uringh is speech In Catilinam on November8 ,6 3B C, or are we interested in how manyp ersons could typically understand speechesgiven by different orators on different occasions during agiven historical period?For the first problem, therewill usually be ahighdegree of modeling uncertainty,because we do not know exactlyhow powerful the voice of this specific person was, what the rhetorical concept of his speech was,orh ow large and how attentive the crowd was on this specific occasion. Hence, we willbeconfronted with ahighmodeling uncertainty leadingtoacorresponding uncertaintyabout the level of the results on speech intelligibility.F or the second problem, the historical scenario itself is described by ar angeo fi nput parameters rather than as ingle model state, and the limits of this rangec an usuallyb ed etermined more preciselyt han the conditionso fo ne specific event.W ec an determine quite preciselyt he rangei n which the sound power of trained speakers lies (e. g. through empirical investigations on present-day subjects),byassumingthatthe size of the crowdcould be something between virtuallyz ero and an audience corresponding to the maximum density of standing persons possiblei napublic place, or by taking into account the fact that the level of attentiono fa na udience at public speeches in terms of the background noise level can again be empiricallydetermined, assuming that the noiselevel producedbyamoderncrowdisnot substantiallydifferent from ac rowdd uring the Roman period.
In order to describe arangeofhistoricalconditions corresponding to astate space of computer models instead of one specific historical occasion, similar strategies to those being used to describearangeo fm odeling uncertainties can be applied, i. e. by calculating the model for the limits of the rangeorbyestimating aprobability distribution for the respective input parametera nd by letting as ampled version of this distributionp ropagate through the simulation in order to calculate the resulting uncertainty at the level of the parameter of interest ( fig.4). In practice, due to the limited information about single events and the empiricallym ore substantiated knowledge about the rangeo fh istorical conditions, there will often be atrade-off between modeling uncertainty and historical specificity. Forthe results that virtual reconstructions mayprovide for adefined rangeo fhistorical events are usuallym uch more reliable and thus scientifically more valuable than thosef or specific historical events.

Reliability and biasI I: The observer
Unlikeamerelynumerical simulation, avirtual historical environment (VHE) encompasses interactive real-time sense data (sometimes also autonomous artificial agents) thata re experienceda nd interpreted by human agents in order to make new knowledge claims about the historical circumstances unders imulation. To legitimate such claims, even in the face of at echnicallyp erfect simulation of historical reality,s cientists employing such as trategyshould always follow established scientific quality criteria for conclusions drawnf rom systematic observations.A fter all, human agents tend to be unreliable, unique and autopoietic measurement instruments. They have constrained attention as well as different cognitive-perceptual capacities and they construct the identity of objects, events and their meaning from the background of their historical culturalu pbringingand embeddedness. Furthermore, in interactive environments, different human agents maybydefinition perform different actions, thereby alsoarriving at idiosyncratic impressions of the very samesimulated worlds.These epistemological problems are well known in research disciplines that have at radition of dealing with subjective field observations, such as psychology, sociologya nd ethnography. Within these disciplines, several methodological answers have been developed to address these challenges. Even the problem of systematically performing social science in simulated virtual realities has been discussed for nearly2 0y ears now (Markham 1998). However,t he type of systematic observation performed within VHEs differs radicallyfrom psychological virtual reality experiments (Veiling et al. 2013) as well as from cyber ethnography (Hallett/Barber 2014)insociologyorcommunication studies in that it is not directed at analyzing the actionso ri mpressions of other human agents within virtual worlds or computer games but at analyzingascholar's own subjective experience of being an actor within asimulated past culturalsetting.Therefore, the methodological approach discussed here mayb ec onsidered an ew form of empirical inquiry that could be termed cyber phenomenology.⁵ Nevertheless, we suggest building on  See Houliez/Gamble (2012)f or ab rief introduction intot he phenomenological approach.
On the Epistemic Potential of VirtualR ealities fort he Historical Sciences methodological techniques developed within the abovementioned fields, since, as we will try to show in the following,t hey help to improvea nd demonstrate the reliability and intersubjective validityo fs ystematic scientific observations performedw ithin VHEs.

Performing (semi-)structured phenomenologicalo bservationsi nV HEs to increase reliability
The most important challengef or cyber phenomenologyf rom a reliability perspective appears to be the selectivity of human perception, regardlessofcultural background and expertise. Attention span,cognitive priming and halo effects (Kochinka 2010), as well as the results of different actions performedi nt he same virtual environment by different actors maylead to very different subjective percepts.Astraightforward methodological approach thati sa blet o' tame' the resultingi diosyncrasy of impressions with al ong history of successi nt he social sciences is performing structured or semi-structured observations (Bryman 2008b). Again, these requireawell-specifiedr esearch question and ac learly specifiedr angeo fp henomena of interest at the outset,which are then used to createanobservation form (or category sheet) that formulates different activities ar esearcher has to perform (e. g. visiting certain places within the VHE) and leavese mpty slots in the table to be filled out during or aftert he experience. These could be used to describet he subjectively experienced mere existence or the experienced qualitieso ri ntensities of certain a-priori expected phenomena.
In the fullys tructuredv ariant,a ll phenomena and their subcategories are alreadyspecified at the outset of the fieldwork, turningthe table form into amultiple-choice questionnaire.I nt he semi-structured version, the categories for phenomena to look out for tend to be rather broad and the subcategoriesa re openended, which leavest heir creation to the field-observer, and thereby introduces an interpretive-integrativestep in the lateranalysis (Gehrau 2002). Regardless of which of the two approacheshas been adopted, duet otheir degree of standardization, both procedures principallya llow the observations to be converted into numerical variables in alater step and thereby enable aquantitative comparison between the impressions of al argers ample of different historical researchers. This pavest he wayf or systematic mathematical reliability checks, as well as for performing scale buildinga nd statistical hypothesis tests (Bryman2 008b). While internal validityand reliability are improved considerablyw hen performing such structured observations,adrawback of these procedures is that they are onlyablet odeal with a-prioriexpected phenomena, therebyimposingadeductive inferential style of inquiry with the typical tradeoff in external validity.Taken together,s tructured observation techniques appear well suited for quasi-experimental historicals tudies using VHEs thatt ry to test specific hypotheses.

Employing open-ended ethnographic techniques in VHEs to increase intersubjective validity
The most important challengefor cyber phenomenologyfrom a validityperspective appears to be the cultural-historicalsituatedness of human perception: every observational act is ladenw ith implicit culturally-specific assumptions that structure the kind of entities, qualities, actors,r elationships and events constructed by the human mind duringp erception (Bogen 2014). This problem even increases when research questions are of an exploratory nature. Therefore, from the standpoint of validity it appears important in these cases to employ a form of inquiry thatmaximizes the intersubjective traceability of subjective interpretations in order to 'tame' perception'sculturalboundedness. Obviously, this is onlypossiblei fn ot the experiencedp henomena alone, but also their subjective interpretations and the waythey werereached are analyzed and comparedqualitatively between different researchers with different backgrounds and areas of expertise. However,t his is not possiblew ith standardized observation forms (Bryman2008a). Hence, when dealingwith more open-endedresearch questions or when expecting apossible cultural-historical bias on the part of the scientific observer,itseems reasonable to accept drawbacks in reliability and to try to improvei ntersubjective validity by drawing on more 'qualitative' ethnographic techniques developed in culturalstudies and ethnographyinstead of performing structured observations (Hallett/Barber 2014). Typical examples would be historical research projects that are not onlyinterested in questions of the mere perceptibility or intensity of phenomena, but also in certain forms of aesthetic, emotional or social judgments. Hence, to ensuret he intersubjectivity of the phenomenological findingsf rom these scenarios,wer ecommend that researchers try to formulate "thick descriptions" (Geertz 1973)o ft heir subjective impressions in the form of extensive written field notes or by employing think-aloud techniques (Bryman2 008a). To increase traceability of this subjective interpretive data, it should be enrichedbylocative data (positions,distances and actual movement in the VHE) and audiovisual-field recordings(screenshots, short movies and audio renderings) which would help other scholars to understand and tracet he personal interpretations and conclusions. Furthermore, researchers performing cyber phenomenologyi nt his ways hould explicitlyd isclose the stocks of historical cultural knowledge that they actively employ whent rying to put themselvesi nto the hermeneutic position of ac ontemporary observer. By triangulating these different forms of data in an interpretive analysis, which maya lso comparet he field notes of as maller number of researchers visiting the same scenario, a grounded theory (Bryant/Charmaz 2011) about previouslyu nknown mechanisms and phenomena in the field mayt hen be developed. Taken together,q ualitative ethnographic techniques appear wells uited for more open-ended, explorative historicalstudies using VHEs that try to arrive at new hypotheses.

Combining structured observations and ethnographic techniques
Obviously, ethnographic techniques and structured observations can also be combined in ac omplementary wayt oa ccommodate for each of their tradeoffs. Taking the research example discussed above, ah istorical scientist could have first ethnographicallye xplored several VHEs with different historical versions of the Forum Romanum to arrive subsequentlya tt he grounded theory that some of the positional changes of the Rostramusthavebeen related to acoustic problems.This could give rise to an additional studyemploying structured observations in the sameV HE in order to test this hypothesis statistically.

10.N ew knowledge based on existing evidence?
Af inal remark will be made on the possiblef uture role of virtual environments and the related observation techniques we envision within an epistemological model of historiographical research. If the empirical material collected by historians is regarded as material traces of historical events that give rise to collective inferences regarding 'common causes' (Tucker 2004), one still has to explain the additional value of virtual environments,when these can onlyb eb ased on evidence thati sa lreadyknown. Every aspect of the example simulationo ft he virtual Forum Romanum discussed in this article, from the forum'sarchitecture and the surrounding buildingst ot he location of the Rostraw as based on previous knowledge in classicala rcheology,a nd the computer model did not bring to light anyn ew historicale vidence.
Nevertheless,wesee threewaysinwhich new knowledge could emerge from virtual or augmented historicalenvironments.First,bysystematicallyrelatingtogether scattered bits of information(about the architecture, the architectural materials used, the location of the Rostra, the text of the speech) within amodel of sound or light propagation, new informationc an be derivedt hat was already embedded in the historicale vidence available but had not been combined before. Thus, the scientific theories about sound or light propagation take on the role of nomicr egularities, which carry "nested information" about the source of information, i. e. the original historicalevents (Tucker 2004,18f.).⁶ This is similar to the application of natural laws about the ageing of materials in order to extrapolatet he original appearance of the materials from the existing remains together with knowledge about the climatic conditions of the past.S ince disciplines such as geology, meteorology, the material sciences or acoustics are currentlyintensively involved with the development of computer models for the dynamic, i. e. time-related, behavior of physical systems,t he historical sciences would do well to exploitthe wealth of nomic regularities inherent in these models.
Second, sincev irtualo ra ugmented environments area blet ol ift the numerical variableso fc omputerm odels to thel evel of sensorys ignals,t hey not only convey more comprehensivei nformationo nq uestions concerning thep erceptuali mpression of historical environments; thesea re also easier to access explanatorily for non-experts in therespective scientific domain. Apredictiono ft he speech intelligibility in the ancient Forum Romanum could, in principle, already be made based on information about the soundp ower of traineds peakers,t he distanceb etween speakera nd audience,t he reverberation time of theplace andthe noiselevel generatedbythe audience. However, thea nalytical relations involved aret hemselvesd erived from psychophysical experiments. As statistical correlations, they have only limiteda ccuracy when predicting the intelligibility of presentations of texts in as pecificl anguage withaspecificr hetorical strategyf roms everal listeningp erspectives. And it is one thingt ok now that the predicted speech transmissioni ndex (STI)a tacertainp lace in thea udienceh as av alue of 0.7a nd another thing to be able to listen to thev irtualizationo ft he speechi nd ifferent places in the audience -at least forl isteners whoa re not familiarw itht he meaningo f STIv alues. Insofar,VHEsh elpt oe xplore thea ctualp erceptualmeaningofc oefficients derived from meren umerical simulations.
And finally,t he integration of knowledge stocks from different academic disciplines, required for the empiricallys ubstantiated design of virtual historic environments,m ight make these af ocal point of interdisciplinary cooperation, wherescientific knowledge from areas as farapart as classicalarcheology, architectural and building history,the history of rhetoric, room acoustics and speech acoustics convergeatone common point.And the history of science is full of exampleso fn ew knowledge emerging from an ew and unusual cooperation between different disciplines.