South African Archaeological Society

Stone Implements as InformationAuthor(s): J. E. ParkingtonReviewed work(s):Source: Goodwin Series, No. 1, The Interpretation of Archaeological Evidence (Jun., 1972), pp.10-20Published by: South African Archaeological SocietyStable URL: http://www.jstor.org/stable/3858089 .Accessed: 05/12/2011 10:08

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact support@jstor.org.

South African Archaeological Society is collaborating with JSTOR to digitize, preserve and extend access toGoodwin Series.

http://www.jstor.org

STONE IMPLEMENTS AS INFORMATION

J. E. Parkington

Department of Archaeology, University ofCape Town

Introduction African prehistorians, like their colleagues else-

where, are constantly striving after more effective ways of describing and interpreting large collections of stone implements. With the introduction of methods of computing and the growing interest in the methodo- logy of science, attention is directed towards the discovery of more 'objective' less 'intuitive' methods. It would be unrealistic to hope for complete objec- tivity even if this were a desirable end, but it is felt that many of the processes of archaeological studies can be rigorized and improved. In a sense there are three stages of archaeological analysis, the selection of relevant data, its manipulation so as to produce interpretable patterns and the interpretation of these in terms of prehistoric behaviour. The first and third stages require the exercise of intuition on the part of the archaeologist and will seemingly always do so. Such intuition is a reputable part of any scientist's methodology (Medawar, 1968). The second stage requires the reorganization and restatement of large quantities of data in terms of regularities thought to be significant, and it is here where more rigorous, probably numerical, methods would appear to have potential. The discussion which follows is an attempt to set before archaeologists working in South Africa some of the writer's thoughts on contemporary methodologies. The comments on attribute cluster analysis reiterate the contribution of Albert C. Spaulding in two early papers (1953, 1960) and their expansion by James Sackett (1966, 1968). Work on numerical methods has been in progress in South Africa since 1957 when Dr. R. J. Mason began to analyse the lithic material from the Cave of Hearths (Mason, 1957, 1962, 1967, 1971). The writer has experimented on attribute cluster analysis in dealing with small collections of stone implements from the south-western Cape (see Parkington & Poggenpoel, 1971).

Expectations It seems logical to preface a discussion of the

analysis of stone implements by setting out precisely what one expects of such analysis in terms of both 'ends' and 'means'. In other words, suggestions as to how implements might best be studied should be evaluated in terms of stated criteria. These criteria would apply not only to the sorts of information ultimately sought (ends) but also to the ways in which the analysis seeks it (means). Perhaps it can be taken as a basic assumption of all systems of analysis that stone implements represent tangible information about past behaviours and that it is the patterned behaviours represented by patterned

]

implements which are the proper study of the analyst. The following are some personal views as to how stone implement analyses might be assessed which arise directly from this assumption.

(a) Clearly, useful analyses of stone implements should highlight patterns of technological behaviour in terms of selection of raw materials, techniques of manufacture, maintenance and utilization and perhaps patterns of discard. In other words, the whole process of birth, life and death through which stone imple? ments pass should leave traces on the surviving specimens and this information should be extracted.

(b) The recognition of technological patterns such as recurring combinations of characteristics and the presence of more or less standardized implements should give access to a further set of behaviours. For example, patterned implements suggest interpreta- tions as to function and thus give clues to behaviours which require stone implements?such as the use of composite tools, the utilization of other raw materials (skinworking, woodworking, boneworking, etc.) or methods of hunting and collecting (arrows, digging- sticks, spears, snares). Here the associations of patterned implements with other sorts of debris (food remains, hearths, sleeping-areas) would certainly be of use.

(c) Information gained by analysis of stone imple? ments should be integrated with other sets of informa? tion into a comprehensive statement on behavioural patterns as represented at particular archaeological occurrences. It may then be possible to proceed to a survey of regional or continental patterns of behavour using a terminology which highlights simi- larities and differences through time and space. It could be stressed that, in studies of hunter-gatherers, the present hierarchy of terms beyond the level of single site analysis is highly lithocentric and tech- nologically oriented. Despite the fact that such com- munities are usually stone-using, it is no more logical to base groupings of assemblages on patterns of technology than on, say, estimations of group size or patterns of exploited resources or regularities of site location. The most effective way to proceed beyond the stage of single site analysis would seem to be to compare the total behavioural patterns at individual situations and discuss the appearance, spread and modification or disappearance of all aspects of behaviour (social and economic as well as technologi? cal) in time and space. The more so since improved chronological techniques make it possible to observe the precise temporal relationships between occur? rences. To rephrase Daniel (1944), prehistoric archaeo? logy has passed through a geologically orientated

0

phase (an 'epoch' model), through an anthropologic- ally orientated phase (a 'culture' model) and on to an historically orientated phase (no names ?). The analy? sis of stone implements must be put into perspective and must provide some, possibly most, but not all, of the information from which the sequence of pre? historic events can be reconstructed.

(d) Analysis of archaeological materials like those in other disciplines must fulfil certain basic require- ments. The methodology of science is a popular subject in contemporary academic literature and archaeology will surely benefit from the application of rigorous standards to its own procedures. Thus Leach (1970) notes that statements which cannot be refuted are of little scientific value. This seems to imply that useful statements must contain within them the wherewithal for their testing and, if necessary, rejection. In order to achieve this, analyses must be stated in explicit terms so that statements for further testing can be generated. As an example of non-explicit, non- testable statements, consider religious pronouncements such as: 'God exists. He has been revealed to me. When you have your relevelation, you will also believe.' By their very nature, revelations are non-communic- able and religious statements such as the foregoing must be accepted on faith or rejected through the lack of it. Analysis of stone implements must state their terms of reference explicitly and proceed by explicit steps to statements which are testable by the con- sideration of fresh data or by a reconsideration of the same data, and which can be rejected if found to be no longer tenable. Otherwise statements about stone implements will degenerate into religious pronounce? ments incapable of being tested and modified but which must be either accepted as gospel or rejected as dogma.

Two points arise from these views; the first con- cerns the 'ends' of stone implement analysis, and the second concerns the 'means'. The first is that the essential focus is the illumination of patterns, which can then be interpreted in terms of behavioural regularities; the second is that the illumination should proceed in an explicit and testable fashion so as to avoid terminological deadlocks and to maximize communication of ideas and interpretations. The next step is to examine analyses presently used in terms of these criteria.

Typologies

How do archaeologists analyse large collections of stone implements? Until recently most analyses were of the intuitive typological kind, the procedures of which were, and are, roughly as follows. The analyst, being accustomed to dealing with specific sorts of collections (from limited locations in time and space) and armed with some experience of the range of implements to be expected is able to divide up implements into a series of 'types' or recurring forms to which names are assigned and the frequencies of which are noted. Assessment of differences and simi-

11

larities of assemblages (and therefore, it is supposed, historical affinity between groups of implement makers) is by a comparison of the presence, absence or frequencies of the selected types. This leads to

'types' of assemblages which form the superstructure of archaeological reconstruction and which are

variously termed 'cultures', 'industries', 'phases', etc. In the past decade many archaeologists have become critical of such procedures and much attention is now directed toward alternative systems of analysis. Some criticisms of this approach are:

(a) The type concept has been applied at the imple? ment level and the cultural level; the former resulting in 'handaxes', 'cleavers', 'choppers', the latter in

'Aurignacian', 'Solutrean', 'Magdelenian'. Many of the implement and cultural type names were coined in the nineteenth or early twentieth centuries when

they proved useful reference points in a discipline still groping in the dark. However, there was a danger that these reference points would become pigeon-holes into which further data would be pushed or at times forced. Thus it is an accident of history that the caves of Aurignac, Solutre and La Madeleine were investi-

gated in the early days of French prehistoric studies and became, as a result, the standard units of com?

parison. It is an historical accident that during the nineteenth century a relatively small number of

implements were recognized as such and assigned names. At the time it was useful to be able to refer new finds to the entities 'handaxe' and 'cleaver' or

'Aurignacian' and 'Solutrean', but we may now ask whether sophisticated analysis should bolster this model or search for an alternative. As early as 1939

Clyde Kluckhohn said: 'The attempt to quantify faulty concepts will only lead to concepts which are more precisely faulty' (Kluckhohn, 1939). The

question is not 'how can we define the terms "handaxe" and "cleaver" ?" but 'are there any patterns of imple? ment manufacture in the assemblages of the lower

palaeolithic to which we would wish to assign names ?' The 'types' may be useful or they may not be, but their utilization leads the analyst away from the search for alternative explanatory models.

(b) A more fundamental criticism of the 'type' concept is the fact that it presupposes the presence of discrete groupings within the data under examination. Intuitive typological analysis has never failed to reveal the presence of 'types', yet further analysis might well suggest that the 'types' are in fact arbi- trarily chosen points on a continuum of change. Moreover, the use of types directly obstructs ways of

observing what might be one of the most interesting trends in prehistory, namely the trend toward greater standardization of implements. If we compared the

tendency for specimens to cluster about 'ideal forms' in assemblages labelled 'Oldowan', 'Acheulian', 'Mousterian', 'Magdelenian' and 'Maglemosian', would we find that the clusterings became tighter as we moved through time? The decision to divide all these assemblages into types and count frequencies

completely ignores the questions of how rigidly the implements are patterned and whether this changes through time. Can 'Oldowan types' be regarded as comparable with 'Maglemosian types1 ?

(c) Implement types often bear unfortunate resemblances to religious pronouncements. They are known or 'revealed' only to the initiated, they are used by the faithful. Consider the statements: 'Hand- axes exist. I have counted their frequency. Under my guidance you will also recognize them.' Typologists often feel that they have tested their concepts by teaching others to use them; they have not. It is only when the criteria of 'handaxe-ness' have been made explicit, and when these criteria have demonstrated the existence of handaxes as distinct from other sorts of implements, that the concept could be regarded as acceptable. 'Handaxes' are in fact a good example, as they were recognized as early as the mid-nineteenth century, yet have never really been defined. Perhaps the only attribute possessed by all 'handaxes' is bifaciality, but this is a characteristic of many other implements as well. Many people have extreme diffi- culty in distinguishing between 'handaxes' and other types such as 'cleavers', 'picks' and even 'cores', and we may well ask whether such distinctions represent the most useful analytical tools.

(d) As regards the meaning of the higher order types such as the 'cultures', 'industries' or 'phases' of archaeological terminology, it must be pointed out that there are many reasons why particular assem- blages differ from one another, any one or more of which may be applicable in each case. There is first the problem of defining how dissimilar two assem- blages must be to be declared the products of different cultures, as clearly there exist no objective standards or criteria for such a decision. More fundamental perhaps is the observation that assemblages may differ because?

(i) they are separated in time, (ii) they are geographically remote,

(iii) they were produced by groups of different technological traditions,

(iv) they represent the residues of quite different activities.

(v) they are the results of quite different excavation strategies on the part of the archaeologist,

(vi) the sites have been subject to quite distinct erosive or destructive forces, or

(vii) any combination of the previous six. With this variety of explanations for inter-

assemblage variability, it may be argued that the grouping of assemblages on the basis of frequency of occurrence of the implement types is an oversimplifi- cation. Are 'cultures' or 'phases' the most effective ways of describing the technological progress of pre? historic man? (For a discussion of the hierarchy of archaeological terms, see Parkington, 1970.)

(e) The search for 'types' of implements and 'types' of assemblages seems to have become an archaeo? logical end in itself, an essential procedure in pre-

1

historic studies especially of the more distant periods of the Stone Age. The 'types' extracted from imple? ment or assemblage analyses have tended to over- shadow what is a more fundamental and interesting question, namely the patterning that has allowed their recognition. The ability to sort out stone implements into 'types' as one would playing-cards into suits is of minor importance compared to the reasons under- lying the tendency for implements to cluster into 'ideal forms'. It is not the group 'handaxes' which is important but, as Plato might have said, 'handaxeness'. It is the patterning of attributes beyond what one would expect by chance, which suggests interpreta- tions and which must be close to the 'concepta' of the implement makers (see Clarke, 1968: 182). A type list would seem to be a dispensable stage in the analysis, since it is then necessary to refocus on the reasons for type existence, i.e. the attributes which have recombined often enough to present a recog- nizable regularity.

To summarize: there are reasons for supposing that the search for discrete 'types' of implement is only moderately successful in terms of the criteria of evaluation suggested. The development of intuitive typological analysis has contributed to a rather rigid lithocentric model for describing the regularities of artefact manufacture and the patterning of assem? blages through time and space. If efficiently imple- mented, it will allow the analyst to make statements about some aspects of artefact-making behaviour but it will tend to mask others. For example, it cannot describe the gradations and continua which are emerging from many prehistoric studies, nor will it allow prehistorians to examine several patterns each cross-cutting the assemblages in different planes. More important, perhaps, there exists the possibility of a 'cart' and a 'carthorse' problem. Rather than recognizing entities (handaxes, Aurignacian) and then defining them, should we not survey the data, seek to illuminate patterns or clusterings of units and assign names where required? The problem is not 'quantitation' but 'entitation' (Clarke, 1968) as stated above. Finally, the incommunicability of 'types' and 'type' boundaries is well known and can perhaps be improved by more explicit systems of implement analysis; more specifically by focusing attention on patterning of attributes rather than the 'type' categories which, in effect, summarize such patterns.

Attributes

Underlying all implement analyses, whether explicitly or implicitly, is the feeling that it is the observable attributes of individual specimens which preserve information and which must therefore be the real focus of attention. The use ofthe phrase 'attribute analysis' in archaeology is not new, but there is still considerable misunderstanding about its meaning. In a sense all implement analyses are 'attribute orientated' but some are more explicitly so, especially since the use of computers and numerical methods

have become prevalent. Perhaps the earliest attempt in archaeological literature to make use of explicit attribute analysis in the examination of stone imple? ments is the 'Statistical Techniques for the Discovery of Artefact Types' by Albert C. Spaulding (1953). His position was clearly stated as follows: 'The arte? fact type is here viewed as a group of artefacts exhibit- ing a consistent assemblage of attributes whose com- bined properties give a characteristic pattern. This implies that, even within a context of quite similar artefacts, classification into types is a process of dis? covery of combinations of attributes favoured by the makers of the artefacts, not an arbitrary procedure of the classifier' (Spaulding, 1953: 305). In a later paper the phrase 'attribute cluster analysis' is coined to describe this procedure, about which he said 'the basic feature is the comparison of the observed count for each attribute combination with the count expected on a hypothesis of attribute independence. These comparisons result in a rating on a continuous scale for each attribute combination along the lines of less than expected, about the same as expected, and more than expected' (Spaulding, 1960: 443). These ratings could then be interpreted as combinations avoided by the makers, those neither avoided nor sought after, and those preferred by the makers. Spaulding added: 'So far as I can see, classification with respect to attribute clusters exploits fully the formal information presented by a collection of artefacts since it inter- relates the entire list of discriminated attributes in terms of both attribute frequency and attribute com? bination frequency' (Spaulding, 1960: 443). It is 'simply a full dress exposition of the reasoning implied in the shorthand statement that type classifications are accomplished by putting together artefacts that look alike' (Spaulding, 1960: 443-4). In addition: 'It is important to note that [attribute] cluster analysis does not create clusters when they are not implied by the empirical data . . . and it will not necessarily assign every artefact to a typological pigeon hole' (Spaulding, 1960: 444).

These views have been utilized and expanded in two papers by Sackett (1966 and 1968) in which attribute cluster analysis is employed in the study of the French Upper Palaeolithic sequence, and more specifically of Aurignacian end-scrapers. Sackett feels that 'the best solution available to this problem [the discovery of patterning amongst implements] entails replacing intuitive type design by an approach known as attribute cluster analysis, which isolates typological categories by means of an explicit quantitative analysis of patterns in which formal attributes segregate non- randomly (that is, cluster) among artefact samples' (Sackett, 1968:73).

These statements by Spaulding and Sackett have been quoted at length since they present clearly the aims and advantages of explicit attribute analysis as compared with implicit 'type' analysis. These advan? tages can be summarized as follows:

1. Patterns or combinations of attributes which occur more than should be expected by chance ought

to represent the residues of patterned technological activities in the past.

2. They should also provide the best potential for interpretation as to function and as the components of implement-orientated activities.

3. The patterns of activities suggested by attribute clusters can then be added to interpretations of other data to go forward to the second stage of archaeo? logical reconstruction, that between assemblages or between sites.

4. Such patterns have been obtained explicitly by testing the tendency for specific attributes to covary and for specific attribute states to combine con- sistently within a given sample of implements.

A suggested procedure might be as follows:

Step One

Choice of relevant attributes and their measure- ment on the sample of implements. The number of attributes and attribute states should initially be as exhaustive as possible, since attributes can be aban- doned and attribute states amalgamated at a later stage in the analysis.

Step Two

A review of the numbers of implements exhibiting the various states of each attribute will now allow some of the characteristics of the sample to be described. Thus, the choice of specific raw materials, the tendency to produce specific sorts of blanks, the overall size tendencies of worked pieces or the con? centration of implements in particular areal or depositional units can be noted.

Step Three

An attempt can now be made to assess the degree to which the attributes are in any way correlated. Following Spaulding (1953, 1960), it is suggested that a simple chi-square test of the frequencies of each combination of two attributes will illustrate signiflcant and non-significant incidences of correlation at chosen levels of significance. (For a full discussion of non- parametric statistics in disciplines such as archaeology, see Seigel, 1965.) The signiflcant and non-significant combinations of attributes can be presented in tabular form.

Step Four

Having illustrated the tendency for some of the attributes to vary in a non-independent fashion, the more interesting examples can be, and must be, followed up. It is only part of the story to know that the size of an implement is in some way related to the raw material used. We wish to know which raw materials were used for which sized pieces.

A closer examination of the contingency table for any particular pair of attributes will demonstrate where the significance of the signiflcant chi-square reading originates. In order to interpret these figures, some policy for the amalgamation of attribute states into super-states is required. One possibility can be

13

suggested. Let us suppose that the following figures were achieved:

The chi-square test has suggested that there is not an independent relationship between variables A and B; it must now be decided which attribute states of each variable are in fact correlated. It is clear both visually and in fact statistically that the distinction into states A2 and A2 has no meaning in terms of variable B. Similarly the distinction into A3 and A4 and into Bx and B2 are not meaningful, and this can be demonstrated by applying chi-square to the more limited sample of say A3 and A4 only (omitting Ax and A2) or Bx and B2 only (omitting B3). Thus super- states can be found consisting of A1+2, B1+2, A3+4, B3 beyond which amalgamation cannot proceed without altering the chi-square readings. It is now possible to state that implements exhibiting superstate A1+2 also tend to exhibit superstate B1+2, and that those exhibit? ing superstate A3+4 tend to exhibit (super-)state B3. This could be rephrased in behavioural terms accord? ing to the nature of the attributes concerned. It must be pointed out here that superstates A1+2 and A3+4 are relevant only to variable B and could be replaced by say A1+2+3 and A4 when other variables are under study. The extent to which this is the case is an interest- ing phenomenon, worthy of further comment if and when it occurs.

Step Five

It should now be possible to describe more para- meters of the implement collection in terms of pre? ferred combinations of superstates. For a complete statement upon how to measure the significance of particular combinations from the chi-square con- tingency table, see Miller (1966: 218, 219). Using the formulae suggested here, all possible combinations of pairs of superstates can be graded, as Spaulding (1960) noted, on a scale ranging from avoidance to selection. However, problems arise when combina? tions of three, four or more superstates are sought. There are two or more alternative strategies here:

1. It is possible to generalize the two-dimensional contingency tables into three or more dimensions and to extract signiflcant combinations as above. However, the problems of computation are enormous and the attenuation of cell sizes on small collections is critical. As a result, it is unlikely that such a procedure would be (a) financially or operationally advisable, or (b) meaningful, though Sackett presents a four-way chi-square contingency table in his 1966 paper (Sackett, 1966: 373).

2. A simpler though, in many ways, cruder method is called 'equation' by Sackett (1966: 369). By this method it is assumed that a multivariate analysis of

superstate combinations can be replaced by a series of bivariate analyses. Thus, if superstate A1+2 is correlated with superstate B1+2, B1+2 with Q and Q with A1+2 it is assumed that all three form a super? state combination of some significance. Whilst some information is lost in these comparisons or 'equations' it does seem a promising technique for isolating com? binations between more than two attributes.

3. Another suggestion, as yet not investigated, is the possibility of attempting a form of cluster analysis amongst superstates from a series of attributes. This suggestion may be open to the criticism levelled by Mathews (1963) at the analysis of Beaker pottery by D. L. Clarke (1962).

Step Six

Some attempt could be made to assess the numbers of implements upon which combinations of super? states tend to appear or be represented. Thus two sites may both exhibit a preferred combination of A1+2 with B1+2 but the combination may be much more common at the one site than at the other, despite the fact that it is an equally significant com? bination at them both. This 'numerical strength' of preferred combination may be interpretable in terms of chronology (combinations may have an ontogeny curve-Clarke, 1968) or of activity (combinations presumably reflect specific activities). At the time of writing no final decision has been made on how to assess such 'numerical strength' (for suggestions, see Sackett, 1966: 366-8).

Step Seven

The parameters of the implement sample resulting from the analysis of individual attribute frequencies, from the preferred combination of pairs (or more) of attributes and from related phenomena can now be interpreted in terms of patterns of prehistoric behaviour and integrated into an overall 'assemblage interpretation'. Data from non-artefactual sources can be included and added to the reconstruction.

Step Eight The regularities of implement manufacture and

use as well as the other aspects of prehistoric life can then be temporally and geographically ordered so as to allow interpretation and reconstruction above the single site level. The appearance, spread and dis- appearance or modification of attributes, attribute clusters and their strengths, numerical or otherwise, can be discussed without recourse to discrete units, unless of course such units are suggested and defined by the data under analysis.

An example?De Hangen

During 1968 a small sample (288) of worked stone implements was excavated from the shallow deposits of the cave De Hangen in the south-western Cape Province of South Africa (see Parkington & Poggen- poel, 1971). This material has been used as a pilot study to test the suggestions made earlier, the results

14

of which are set out below. The figures quoted are never percentages unless specifically described as such.

Step One

The information collected from the implement sample in terms of chosen attributes and states is to be found in the Appendix.

Step Two

The overall characteristics of the De Hangen assemblage are its small size (141 implements are less than 1 gm, and a further 99 are between 1 gm and 5 gm) and the predominant use of quartz as raw material (155 are of quartz). Siliceous stones such as silcrete (41) and chalcedony (42) were used in about equal proportions and lesser use was made of indu- rated shale, often known as lydianite (14). Quartzite implements are similarly uncommon (21). The distri? bution of worked pieces is roughly proportional to the volume of unit represented and no signiflcant concentrations areally or depositionally are detectable. Leaving aside broken pieces for obvious reasons, the position of retouch seems to fall into either the end, the side or the diagonal (E/S) positions as described in the Appendix. Few specimens exhibit combinations of these positions (24) as compared with those with only one position (128). This working may take the form of grinding (19), vertical unifacial retouch (46), less steep to positively shallow retouch (92), crushing and splintering bifacially represented (68) and a num-

ber of instances of combinations of the above (56). On the majority of pieces not broken the working is confined to less than one-third of the perimeter (100) although on some it is between one- and two-thirds (47) and very occasionally more (2). The sorts of pieces chosen for use as implements may be easily recognizable as flakes (190), possible flakes but not exhibiting either bulb or recognizable bulbar frac- ture (57), cores or chunks (22), slabs (1) or pebbles (18). Worked edges may be straight (61), concave (34), convex (123) or some combination of the three (29) and some were difficult to assign to a particular state (irregular = 23). Ofthe whole implements, most (111) were less than twice as long as broad, and a smaller number (36) were more than twice as long as broad, with the pieces orientated as set out in the Appendix.

Step Three

Table I illustrates the extent to which each pair of attributes could be considered mdependent. Daggers represent non-significant chi-square readings at p = 0,05 and therefore suggest attribute indepen- dence; asterisks mark significant chi-square readings and suggest non-independence. It follows from this table that attributes 6 and 3 show complete inde- pendence and will thus not repay further investiga? tion. Perhaps the most interesting aspect of these results is the interrelationship of variables 7, 8, 9, 10, 11 and 12 as suggested by the solid triangle of asterisks in the bottom right section of the table.

Table 1

weight

broken

mastic

material

deposit

area

retouch position

retouch nature

retouch extent

type of piece

shape of edge

shape of piece

0,05 = signiflcant not signiflcant

15

Variables 1, 2 and 4 are independent of some but not all of the other variables.

Step Four

The creation of superstates was undertaken as described above in the cases of attribute combina? tions suggested to be more common than expected by chance (see Table 1), and for the sake of complete- ness in the others too.

concave -> E or S med/shallow -? convex crushing -> straight overlap -> conc/conv crushing -> ? med/shallow -? ? | -> med/shallow not flake -> less than J core -> less than J i to f -> flakes straight -? less than J concave -? less than i i to f -> convex less than i -> ? cores -> straight concave -> flakes convex -> flakes | -> flakes not flake -> ? cores -> ? | -* convex straight -> Q concave -> Q

Quantitatively, the number of times a superstate appears on the left-hand side of the above relation? ships is an indication of its tendency to correlate with other superstates. The breakdown is as follows: core (6 times); |, vertical, overlap, not flake, E/S, concave (5 times); i to f, small, med/shallow, crushing (4 times); straight, minute (3 times); CSL, E or S (twice); less than J, convex, quartz (once); flake (never); ? (never).

This order corresponds to the extent to which superstates in one attribute tend to be combined repeatedly with specific superstates in other attributes. Thus the superstate 'core' is often combined with other superstates whilst the superstate 'flake' is more evenly distributed through the alternative superstates of other attributes. Making use of the equation technique suggested by Sackett and with recognition of the relationships of paired superstates presented above, the following sets of superstates have been achieved: | -> minute E or S

-> flake

16

^EorS -?flake -> convex

-? flake -> -> convex -> flake

A combination of vertical/minute/E or S/J to f/flake/convex is apparent.

D At the point marked by an asterisk (*) this rela-

tionship connects with another but the suggestion is that a combination of flake/quartz/E/S/crush/splint/ straight/less than ?/ ? is represented. med/shallow -> E or S -? flake

-> convex ^ flake -> convex -> flake -*?

The combination med/shallow/E or S/flake/con- vex/ ? is suggested.

By tracing the other superstates in this way no fresh combinations are obtained, although the quartzite/pebble/grinding/large combination has been left out since it was an obvious result after the indivi? dual attributes had been measured. It can thus be added to those listed immediately above.

Step Six

It is a simple process to list the number of occasions upon which the superstates of a suggested combina? tion occur together on specific implements. However,

17

since it is the trends rather than the ideal types which are of interest, it may be convenient to list the number of implements with, say, any five of six or any four of six superstates, thus not requiring implements to possess all of the superstates, merely the bulk of the suggested combination.

Step Seven

The parameters ofthe De Hangen implement sample must have become fairly clear from what has pre- ceded. The collection is dominated by flake imple? ments of small (less than 5 gm) size, over half of which were manufactures from quartz. Secondary working is predominantly unifacial trimming at a variable angle from the bulbar surface, although the crushing and splintering on a minority of pieces is bifacially represented and perhaps results from utilization rather than retouch as such. Pieces with recognizable bulbar surfaces tend to be flakes rather than blades, though a number of pieces show no characteristics of conchoidal fracture, resulting no doubt from the characteristics of the raw materials used.

Implements made from quartzite pebbles tend to be more common in the deposits of the pit than elsewhere, but otherwise there seems to be little correlation between areal and depositional associa- tions and formal characteristics. The presence of fractured surfaces and traces of mastic seem to be independent of the other characteristics. However, the attributes of raw material, weight and retouch do not appear to be independent. On the contrary, there is good evidence that the prehistoric implement makers consistently selected particular combinations of attribute states and avoided or were indifferent to others. The combinations selected could be described as follows:

Preferred superstate combination 1. Fist-sized river pebbles of quartzite were selected and used as grinding stones; they are ground or polished but never chipped. It seems they may have been used in connection with food preparation since such implements tend to occur in the deposits of the pit in association with hearths and food debris. The lower grinding surfaces at De Hangen are not mobile slabs but are small smoothed areas on the bedrock platform at the mouth of the cave. The proximity of the pit to these areas may also be relevant to their apparent concentration here.

Preferred superstate combination 2. Elongated but very small (under 1 gm) flakes were retouched along one or both long sides to form a slightly convex edge almost vertical to the bulbar surface from which the retouch was effected. This form of unifacial blunting occurs on more than a third of the implement peri- meter and no specific raw material appears to have been chosen. The very small size of these implements suggests that they were hafted and indeed mastic is present upon some. They may well have been the tips of composite projectiles.

Preferred superstate combination 3. Unifacial

secondary working at an angle medium or acute to the bulbar surface, rarely if ever vertical, was applied to small flakes up to 5 gm in weight. This results in a smooth convex working edge around some portion of the perimeter of the implement and a length to breadth ratio of roughly one to one. Once again it seems impossible to associate any specific raw material with this combination. The presence of mastic on some implements again suggests that they were mounted into a handle and perhaps used as scrapers.

Preferred superstate combination 4. Flakes of chalcedony, silcrete or lydianite were chosen for the manufacture of implements often exhibiting a hollow or notched worked edge. Flakes between 5 and 20 gm were intensively worked along a small section of their perimeter, usually less than one-third, by the removal of a large number of tiny flakes often overlapping one another. The resultant worked edge is steep or even overhangs the bulbar surface from which the flakes were struck, and on most the outline of the edge is concave. Mastic again suggests that these implements were hafted and the hollow profile may be compared with that of woodworking tools such as spokeshaves.

Preferred superstate combination 5. Quartz has a brittle fracture and often fails to exhibit bulbar characteristics even when it has been worked. Shatter pieces of quartz without recognizable bulbs of per- cussion were selected and used in such a way as to produce bifacially represented crushing and splintering along small sections of their perimeters. This utiliza- tion, if such it is, appears on pieces with a length to breadth ratio of roughly one to one and with its straight profile gives them a rather angular or poly- gonal outline. With the length of piece defined as the maximum linear measurement on the plan view, the utilized edges are generally adjacent and diagonal to the length. It is difficult to suppose how such pieces may have been used, but they may have been hafted, as were most of the non-quartzite implements.

Preferred superstate combination 6. Small elongated flakes, less than 1 gm in weight, were selected and worked unifacially from the bulbar surface. The working takes the form of medium or shallow flaking never vertical to the bulbar surface and forms a shallow convex profile along one of the long sides of the implement. The working of these pieces is similar to that on preferred superstate combination 3, whilst the overall size and shape of pieces resembles pre? ferred superstate combination 2.

Further preferred combinations of superstates may emerge when collections of implements from similar sites have been added to the sample. Thus there are implements from De Hangen which look potentially 'ideal forms' but more examples are needed to show significant figures for the association of attribute states. It is hoped that further research will enable a more quantitative expression of the significance and numerical strength of the combinations suggested above.

Step Eight Too little work has been completed for any regional

let alone larger scale statements to be attempted. However, it is possible to pose a set of questions which might be answerable if not answered in the next few years.

1. To what extent are the parameters of the De Hangen implement sample repeated at contemporary sites in the south-western Cape?

2. Are differences between contemporary sites a reflection of variations in activity, or of available raw materials, or of different implement-making traditions ?

3. Is it possible to demonstrate the appearance and disappearance of implement making habits through time? Can we say at what stage prehistoric man began to put together particular characteristics repeatedly on sets of implements ?

4. Will more extensive observations of preferred combinations of superstates with relation to other cultural debris allow us to interpret such combina? tions in terms of prehistoric usages ?

Conclusion In his book The Structure of Scientific Revolutions

(Kuhn, 1962) Thomas Kuhn illustrates that scientific disciplines change not by the steady inclusion of further theory or knowledge but in a series of explo- sions or revolutions from one model or paradigm to another. The process is roughly as follows. In any discipline a paradigm is set up when the majority of researchers agree on the sorts of problems and questions which it is legitimate to investigate and the ways in which it is legitimate to investigate them. Thus the paradigm specifies both ends and means. However, in pursuing these questions, cracks begin to appear in the structure of the paradigm. Legitimate problems are solved but suggest further problems not strictly within the paradigm, or the increase of knowledge becomes no longer wholly manageable within the limits of the paradigm. Gradually the pressure on the paradigm is increased until the struc? ture collapses and from the rubble a new paradigm is set up. The new paradigm is able to synthesize the increased amount of knowledge and it suggests new questions and perhaps sets up new methodologies to answer them. However, if experience is any guide, it too will be replaced when it can no longer be patched up. Ironically it is the attempt to explore the full limits of one paradigm which leads to its replacement by another.

Perhaps this model is applicable in the field of archaeology, and more specifically in the search for archaeological units amongst implements and assem? blages. In the late nineteenth century a paradigm was set up whereby groups of similar implements were arranged into 'types' and groups of similar assem? blages into epochs (later cultures). During the twen- tieth century the paradigm was improved by a succession of typologists, mainly by increasing the number of identifled groups of implements within

assemblages and by further subdivision of the 'cul- tural units'. More recently various authors have attempted to apply statistical methods to the recogni? tion of types of implement and assemblage. Ironically, the experience of many researchers with numerical methods has led them to reject the paradigm of 'type construction' and pursue a modified paradigm, for example that of 'attribute cluster analysis'.

Whether more explicit attribute-orientated analysis will provide prehistorians with the sorts of informa? tion they would like, only future research will tell. Problems have appeared and will continue to do so, but they may be soluble. Perhaps the real test will be the extent to which new methods of description and analysis prove to be communicable amongst researchers. In 1965 (Inskeep, 1967) Ray Inskeep suggested that the majority of problems facing pre? historians dealing with the Late Stone Age of southern Africa stem from the use of different terms by different workers and the difficulties of deciding precisely what workers mean by the terms used, even when illustra- tions are included. The usefulness of methods of analysis must surely be measured in terms of their success in solving these twin problems.

Acknowledgement I should like to thank Geoff Brundrit and Lee Flax,

mathematicians from the University of Cape Town, for listening to the problems and contributing to the solutions.

each square of the excavation grid was taken as a state, but as no concen? trations were recovered the coding is not included here (for a description of the excavation and the material recovered, see Parkington & Poggen- poel, 1971).

Attribute States Frequency

Nature (of retouch)

Extent (of retouch

Type (of piece selected for use)

Edge (shape of retouched edge)

The position of the retouch was recorded as suggested in the accompanying figure. E-E is the longest linear dimension of the piece and S-S bisects this at right angles

E S .... E/S E &S .. E & E/S S & E/S other combinations and broken

specimens 1. polishing or grinding 2. vertical retouch. 3. approx. 45? retouch 4. shallow retouch 5. crushing and splintering 6. overlapping retouch (step flaking ?) 7. combinations of 2, 3 & 4.. 8. combinations of 5 with 2, 3 or 4.. 9. data not taken .

The proportion of retouched perimeter is recorded less than ? is retouched less than |, more than i less than |, more than \ more than f is retouched three-dimensional working data not taken, includes broken pieces

definite flake .. possible flake .. impossible to diagnose as flake core pebble slab

straight concave.. convex irregular straight and concave straight and convex concave and convex other?largely ground or polished

facets

12 73 43 18 2 2

138

19 19 49 43 68 27 50 6 7

100 39 8 2

19 124

190 21 36 22

61 34

123 23 5

17 7

18:

Shape Specimens are orientated so as to give (of specimen a minimum breadth measurement as a whole) length equal to or smaller than ? 111

twice breadth _ length greater than twice breadth | 36

data not taken, specimens

includes broken 141

In the analysis of attribute state combinations,. meaningless states such as 'data not taken', 'other', 'weight not taken' or 'no clear association' were ignored since combinations including these states would not be useful for interpretation. In addition, the quartzite/pebble/grinding or polishing/larger than 50 gm combination was usually ignored since it was obviously contributing considerably toward the significant chi-square readings and I was interested rather more in the other contributions.

1 would not suggest that this attribute list is either exhaustive or objective, but it does provide some explicit ordering of the information from the De Hangen implement sample. As further collections are excavated and analysed the information will be pooled and the attribute list amplifled or modified, the combination or parameters increased and the possibilities for reconstruction and extrapolation imnroved.

References

Clarke, D. L. 1962. Matrix analysis and archaeology with particular reference to British beaker pottery. Proc. Prehist. Soc. 27.

Clarke, D. L. 1968. Analyticalarchaeology. London: Methuen.

Daniel, G. E. 1950.100 years of archaeology. London: Duckworth.

Inskeep, R. R. 1967. The Late Stone Age of southern Africa: in W. W. Bishop and J. D. Clark (eds) Background to evolution in Africa. Chicago: University Press.

Kluckhohn, C. 1939. On certain recent applications of association coefficients to ethnological data. American Anthropologist 41: 345-377.

Kuhn, T. S. 1962. The structure of scientific revolutions. Chicago: University Press.

Larson, H. J. 1969. Introduction to probability theory and statistical inference. New York: Wiley.

Leach, E. R. 1970. South African Outlook, December. Mason, R. J. 1957. The Transvaal Middle Stone Age

and statistical analysis. S. Afr. Archaeol. Bull. 12 (48): 119-143

Mason, R. J. 1962. Prehistory of the Transvaal. Johannesburg: Witwatersrand U.P.

Mason, R. J. 1967. Analytical procedures in the

Earlier and Middle Stone Age cultures in southern Africa: in W. W. Bishop and J. D. Clark (eds) Background to evolution in Africa. Chicago: University Press.

Mason, R. J. 1971. Multivariate analysis ofthe Cave of Hearths Middle Stone Age artefact assemblages. Johannesburg: Univ. Witwatersrand Dept. Archaeo? logy Occ. Papers No. 7.

Mathews, J. 1963. Application of matrix analysis to archaeological problems. Nature 98 (4884): 930ff.

Medawar, P. B. 1968. Induction and intuition in scientific thought. London: Methuen.

Miller, R. G. 1966. Simultaneous statistical inference. New York: McGraw-Hill.

Parkington, J. E. 1970. Review of 'Background to evolution in Africa'. S. Afr. Archaeol. Bull. 25 (97): 48-50.

Parkington, J. E. & Poggenpoel, C. 1971. Excava- tions at De Hangen 1968. S. Afr. Archaeol. Bull. 26(101 & 102): 3-36.

Sackett, J. R. 1966. Quantitative analysis of Upper Paleolithic stone tools. American Anthropologist 68 (2): 356-394.

Sackett, J. R. 1968. Method and theory of Upper Paleolithic archaeology in south-western France: in S. R. Binford and L. R. Binford (eds) New perspectives in archaeology. Chicago: Aldine.

Siegel, S. 1956. Nonparametric statistics for the behavioural sciences. New York: McGraw-Hill.

Spaulding, A. C. 1953. Statistical techniques for the discovery of artefact types. American Antiquity 18: 305-313.

Spaulding, A. C. 1960. The dimensions of archaeo? logy: in G. E. Dole and R. L. Carneiro (eds) Essays in the science of culture: in honour of Leslie A. White. New York: Thomas Y. Crowell.

20