Gatzarria Cave is located in the western part of the Pyrenees Mountains, in the Soule region (Pyrénées-Atlantiques; Fig. 1). It opens on the northeastern slope of Mont Hargagne, to the east of the Arbailles Massif and overlooks the Saison Valley at an altitude of 270 m above sea level. The cave had been known to the inhabitants of the region for a long time, but the archaeological potential of the site was only revealed in 1950 by P. Boucher. Between 1951 and 1953, P. Boucher, P. Bouillon, F. Bordes, and G.Laplace carried out test pits. In 1956, Georges Laplace began the first excavations of the site, which were interrupted in 1957. They resumed in 1961 and continued until 1976, under the direction of Georges Laplace. Like for the site of Olha II, the “Laplace-Méroc” excavation method was applied (Laplace and Méroc, 1954). However, the recent revision of the collection showed that only large-sized remains, mainly cores and retouched tools, were recorded in three dimensions during these excavations. The rest of the remains were gathered per ninth of a square metre over depths of 5 or 10 cm. Several articles have been published in the past on the characterization of the different Middle and Upper Palaeolithic lithic industries of this cave using the analytical typological approach (Laplace, 1966, Laplace and Sáenz de Buruaga, 2002–2003 and Sáenz de Buruaga, 1991), and on faunal (Lavaud, 1980) or sedimentary studies (Lévêque, 1966).

Several attempts were also recently made at dating the Middle Palaeolithic levels. Four ¹⁴C dates from the most recent level, Cj, place it in the MIS 3, between 44 and 47 ka BP (Ready and Morin, 2013). In contrast, the ¹⁴C results for level Cjr only yielded minimum ages (> 47 400 and > 50 300 BP, Barshay-Szmidt et al., 2012). We can thus consider that the age of the deposits from the base of the stratigraphy of Gatzarria is close or older than the limit of the scope of the ¹⁴C method.

In order to carry out a critical revision of the Middle Palaeolithic lithic assemblage from the Gatzarria archaeo-sequence, it was first of all necessary to carry out distribution and homogeneity tests for the remains in the different excavated sectors.

Indeed, level Cj was excavated over a large surface (44 m²), but the excavation of the underlying levels was confined to test pits over a surface of 1 m² in the entrance and 3 m² in the second half of the cavity.

An earlier study already pointed out differences in the preservation of archaeological occupations in the entrance and the back of the cave for levels attributed to the Protoaurignacian (Barshay-Szmidt et al., 2012; Eizenberg, unpublished). In this way, the results of the taphonomic analysis influenced the sampling of remains in order to characterize the assemblages.

This aspect of the study takes several parameters into consideration (Fig. 2). •

The distribution of the density of remains more than 2 cm long per spit was projected onto a sagittal section. Square 3F and square 6E (which represents the most deeply excavated sector in the second half of the cavity) were selected. The three coordinates were available for some of the remains, but most of them had been gathered by sub-squares of 33 × 33 cm, over depths of 5 or 10 cm. This projection generates the density of remains according to their relative depth, which indicates whether remains occur in concentrations, or whether sterile zones can be identified in the stratigraphy, regardless of sedimentary observations. The main aim here was to test correlations between these two types of data.

In spite of the easily identifiable overall natural dip of the layers from the entrance towards the back of the cavity, which is visible on the surface of the natural ground, Georges Laplace's excavations followed plane surfaces. Therefore, all the assemblages defined during his excavations are probably partially truncated or at least mixed up with their interfaces. The analysis of the distribution of the density of remains can nevertheless identify the presence and morphology of layers of relatively abundant remains at different levels of the stratigraphy, as well as their continuity between the entrance (square 3F, zone A) and the rear sector of the cavity (squares 6D, 6E, 7E; zone B). The integration of the granulometric data and the distribution of the fine fraction can also provide important taphonomic information (Bertran et al., 2012). However, the intensity of post-depositional processes, linked to runoff in particular, has already been discussed here and we have already suggested that these remains are not in primary position (Deschamps and Flas, in press). Moreover, it is impossible to check the sieving protocol implemented during Laplace's excavations, which could bias our interpretations, and thus the interest of this type of analysis in this context must be put in perspective. We thus counted all the lithic remains longer than 2 cm, by spit and by sub-square. In order to represent these quantities on a sagittal section, we then combined the sub-squares on the sagittal sections. In this way, sub-squares 1, 2, 3; 4, 5, 6, and 7, 8, 9 were grouped together (Fig. 2.3).

These analyses enabled us to show that the quantities of remains at the interface between Upper and Middle Palaeolithic levels are not very dense. However, remains are slightly more abundant in layer Cj. In 6E, layer Cjm is an irregular layer with an intense manganese deposit. As previously discussed, level Cjm is not an archaeological level in its own right, but probably formed at the interface of two layers by partially reworking them.

In the two test squares, the density of remains then clearly intensifies in both sectors. Georges Laplace named this assemblage Cjr, but the study of the excavation notebooks enabled us to find the trace of a layer named Cjgr (red-grey-yellow layer) in the upper part of layer Cjr. This Cjgr layer was identified and described in the excavation notebooks, but was not retained in the theoretical stratigraphy published later (Laplace, 1971), as it was considered as a local occurrence affecting Cjr. Therefore, all of these remains were grouped together in assemblage Cjr.

In square 3F, on the other hand, there is no such distinction and this level is called Cjr. But the excavation in 3F stopped after two spits into Cjr layer, and it was thus not possible to assess whether a similar subdivision could be made in square 3F between Cjgr and Cjr.

Subsequently, Laplace decided not to conserve the Cjgr distinction and to group everything under Cjr. However, this subdivision appears to be more significant than a simple sedimentary variation, as the density of remains also indicates a difference between the spits initially attributed to Cjgr and those attributed to Cjr. The results of other analyses also support this hypothesis (cf. below).

The following spits are related to layer Cr, which was only attained by Georges Laplace's excavations in squares 6E and 6D. He described this level as containing few remains, which is also apparent in Fig. 2.4. A subdivision can also be added in assemblage Cr between a first level containing a low density of remains, and a second sub-sterile level from spit 24 onwards, where even sieved elements from the fine fraction are absent (personal observations). •

Secondly, the distribution of rolled remains enables us to show whether post-depositional actions linked to runoff are homogeneous, or whether they are differentially distributed. This type of alteration provides information on the sectors exposed to runoff.

Based on this projection (Fig. 3.2) and in spite of the fact that this information is partial, as it only takes into consideration the recorded remains, it appears that there is a concentration of rolled remains in the sector of the rear test pit of the cavity. In contrast, rolled remains are sparse in the entrance zone. This could be due to: •

remains falling from a vertical alignment and getting mixed up with the remains of the archaeological levels. The latter would then have undergone mass transport towards the interior of the cave, incorporating these rolled remains.

In any case, the rear sector of the cavity appears to have undergone several post-depositional actions, which are clearly less visible in the entrance zone. It has yet to be determined if this sector is relatively intact or if it should be excluded from future analyses. •

In order to determine this question, we tested the refits and conjoins in the sequence (Fig. 3.2–4). The limited excavated surface for these levels (4m² in two distinct zones) was a clear obstacle for refits, but the rare refits and conjoins nonetheless provide crucial information on the conservation of archaeological levels and the correspondence between the square of the entrance (3F) and the sector of the deep test pit (6D/E–7E). The large quantity of remains and the homogeneity of the dominant raw material, a bluish quartzite, represent a further obstacle for this exercise. Furthermore, some of the refitted remains were not plotted during the excavation and the average coordinates of the spit from which they came were attributed to them.

In spite of these difficulties, several artefacts were refitted (Fig. 3.3–4), providing important information on the integrity of the levels. Altogether, 30 remains were grouped into 14 refit units, comprising 9 conjoins and 5 refits. Most of them were on a sub-horizontal axis and a short distance apart (in the same square). Only layer Cj was excavated over a large surface. Two refits identified in this layer are further away from each other (over a meter). However, we observe that these two refits are still sub-horizontal and follow the natural dip of the layer. This indicates that the remains from Cj underwent transport over a short distance, from the entrance towards the back of the cavity.

Six conjoins and refits come from the deep test pit, between squares 6E and 6D in levels Cjgr/Cjr. Their position shows how complex it is to identify clear differences at this level of the stratigraphy. A clearer distinction emerges on the frontal projection of these refits (Fig. 3.4), where two horizons of separate refits seem to occur.

Finally, two refits come from the deepest level Cr, which was only excavated in squares 6D and E. These two refits are at the same altitude and are about 50 cm lower than the others, implying that a relatively better preservation of this level at this altitude. This suggestion is backed up by the distribution of weathered artefacts (Fig. 3.1), which are absent from the bottom of the test pit where the remains display fresh or slightly altered surfaces.

The taphonomic study showed that the aspect of the remains from the entrance was different than those from the second half of the cavity, indicating that the rear sector was exposed to more severe post-depositional action. In addition, our first revision of the Laplace collections show differences in the typological characteristics of layer Cjr between square 3F and squares 6D/E–7E. Indeed, Quina-type scrapers were present in squares 6D/E–7E, whereas they were absent in square 3F. However, it was not possible to clearly identify sedimentary and archaeological correlations between the two sectors, as square 3F was not excavated down to the substratum.

This concentration of remains with Quina-type retouch in the test pit could be due to the differential distribution of activity zones in layer Cjr, or these remains could also correspond to another archaeological level, different from that defined as belonging to Cjr in square 3F.

In light of post-depositional actions, it is unlikely that the distribution of anthropogenic activities was preserved. In the second hypothesis, the absence of artefacts with Quina-type retouch in square 3F, where the excavation did not continue to the base of the stratigraphic sequence, is consistent with the better preservation and greater thickness of the sedimentary deposits in the entrance, as also observed by L. Eizenberg.

In order to test this hypothesis, we made projections of the most discriminating typological elements in the Mousterian sequence (Deschamps, 2014 and Deschamps and Flas, 2017). We selected all the tools plotted in three dimensions and divided them into five main typological groups. It is important to recall here that only some of the cores and retouched tools were plotted during excavations. These five groups comprise: “ubiquitous” tools (1), including flakes with partial retouch and lateral, transverse, double or multiple scrapers with simple retouch, generally found in most of Middle Palaeolithic lithic technocomplexes. They can be differentiated from scrapers with scalar Quina (2) and demi-Quina (3) retouch, which generally attest to specific techno-economic and functional strategies. The notches and denticulates (4) comprise another group and the macro-tools (flake cleavers, 5) form the last one (Fig. 4.1–2).

These projections brought to light a differential vertical distribution of these categories. All the Quina and demi-Quina scrapers, with one exception, are concentrated at the base of the stratigraphy at altitudes ranging between < −170 and −200 cm. Conversely, there are few denticulates in these lower levels. The group of denticulates is concentrated above < −170 and persists until the end of the sequence, attributed to the Middle Palaeolithic. They are thus present in the upper part of layer Cjr (hereafter renamed Cjgr) and in layer Cj. Finally, cleavers are rare and are only present in layer Cjgr.

This reinterpretation of the archaeo-sequence in the test pit enabled us to attempt to position the industry from layer Cjr in 3F in relation to this new stratigraphic sequence. Thus, the typological composition of the industry from square 3F at altitudes ranging between < −135 and −155 cm, and the general dip of the layer, correspond to level Cjgr identified in the test pit where scrapers are the most abundant tools, followed by denticulates, but there are also two flake cleavers and a large non-retouched blank in ophite in the same altitudinal horizon.

The correlation of all these taphonomic, typo-technological and spatial distribution data enabled us to distinguish several coherent levels and to propose a new breakdown of the archaeo-sequence attributed to the Middle Palaeolithic of Gatzarria (Fig. 4.3). The cross-referencing of data concerning the density of the remains, refits and the spatial distribution of typological attributes shows the same interruption in layer Cjr in the test pit. The delimitation between these two assemblages is not clear, but the transversal and sagittal projections identify this limit at a depth of < −170/–175 cm for squares 6D/E.

The fieldwork notebooks also show that the Cjr assemblage was initially divided into two phases, with an upper part named Cjgr in the test pit, with altitudes corresponding to the typological interruption observed on our projections. In squares 6D/E–7E, Laplace's Cjr layer thus corresponds in reality to two distinct archaeological levels: a first one that we rename Cjr-base, above which is level Cjgr.

There are variations in the raw materials used in the different levels attributed to the Middle Palaeolithic.

Quartzites are predominant throughout the sequence (between 78 and 84%; Fig. 5). They are available in nearby river alluvions, in particular in the Saison, which is about 2 km from the site. Samples from the Saison and its main tributaries are currently under study, in order to characterize the different types of Pyrenean rocks available in the local environment of the site and to evaluate the degree of qualitative raw material selection (Deschamps et al., in press and Minet et al., 2019).

The flint is mainly allochthonous, apart from the nearest known source, Iholdy flint, which is about 10 km from the site. The different types of flint identified at Gatzarria show that a considerable variety of regional sources were used, in addition to the Bidache type Flysch flint. These include the deposits of Hibarette to the east, those of Tercis and Chalosse towards the north and Urbasa and Treviño to the south (op. cit.). This implies that human groups had extensive knowledge of a vast regional territory and connections with the Ebro Valley, which also infers that they may have crossed the Pyrenees through valleys and passes in the inland Basque Country.

These flints form the second most frequent category after quartzite, particularly in Cjr-base, where they represent 25% of the remains. Conversely, they are relatively rare in Cj where a local rock, mudstone, was widely used. In all levels, flint was generally imported from distances of 40 to 50 km in the form of flakes and retouched tools (cf. below).

All the other rocks were relatively rarely used, such as ophite, which only represents 0.5% in level Cjgr-3F (where it was most widespread), or quartz, which represents 2.4% in level Cjr-base. The other identified rocks were only marginally used and have been combined in the other/undetermined category (cinerite, lydite, sandstone, indeterminate volcanic rocks, jasper, etc.).

The technological classification of whole flakes also shows variations between the assemblages. In Cj, we observe Levallois production in quartzite (Fig. 6.1, Fig. 7), whereas Levallois blanks in quartzite are practically absent elsewhere in the sequence (Fig. 6.2–5). However, Discoid debitage is also present in this level along with a significant number of pseudo-Levallois points. It is possible that two operative chains coexisted in this level or that there is a palimpsest effect in Cj. The ongoing study of the upper part of level Cj should clarify this hypothesis in the near future.

The industry from layer Cjgr is characterized by dominant Discoid debitage on quartzite (Deschamps, 2014 and Deschamps, 2017; Fig. 8). Most of the blanks from cores are thick flakes with a cortical upper surface. Thus, cortical surfaces are generally directly used as striking platforms. This observation is also backed up by an increase in Kombewa-type flakes in this level (Fig. 6.2–3). Unlike in the overlying level, elongated and Levallois products are rare. Several shaping flakes are present. We also note an increase in products linked to toolkit management. This level also contains a small percentage of flake cleavers (Fig. 9). Flint debitage is rare and flint cores appear to be related to Levallois management, but it is difficult at times to understand production aims as cores were only discarded when highly exhausted (Deschamps, 2014, p. 375).

The limits between the industry from layer Cjr-base and the immediately overlying layer Cjgr are rather unclear. We excluded the remains issued from clearing the interface of these two layers in order to limit possible intrusions. However, it is difficult to eliminate the presence of other disturbances, such as the burrows mentioned at the back of the cave in fieldwork notebooks, and residual disturbances probably subsist between these two levels.

In spite of these biases, the preliminary study of the industry from this level enabled us to identify predominantly Discoid debitage (21 Discoid cores). When blanks could be identified, we observed again that debitage was generally carried out on the lower surfaces of flakes. Four quartzite cores seem to correspond to a Quina-type debitage. They are organized into two secant surfaces from which unipolar sequences were detached with an inward motion (Fig. 10.1). Two flint cores also correspond to Levallois debitage (Fig. 10.2–3). Most of the retouched tools are in flint, even though flint cores are rare (Fig. 9.1). Thus, considerable spatio-temporal fragmentation of the flint operative chains is perceptible.

The debitage methods indicate the coexistence of different operative chains. However, in light of the above-mentioned taphonomic problems in this level, it is difficult to determine whether rare elements derive from stratigraphic intrusions. The coexistence of several similar operative chains was also described for the Quina-type Mousterian in several sites in the Cantabrian region (Carrión et al., 2008 and Lazuén, 2012).

Finally, layer Cr presents clearer differences. In terms of the density of remains, a general rarefaction of remains was identified below the concentration attributed to Cjr-base, then with a less dense layer of remains at a depth of < –220 cm. This assemblage was only identified in a 2-m² test pit, but the absence of rolled remains and the presence of two sub-horizontal refits at the same altitude imply that it is intact (Fig. 3).

It is difficult to identify the dominant debitage system due to the paucity of cores (n = 5). Flake production is characterized by elongated products and Levallois debitage. Retouch and resharpening activities are also well represented. On the other hand, shaping flakes totally disappear in this layer.

If we take into consideration the proportions of raw materials used for the toolkit, we observe clearer variations between levels (Fig. 9.1). Generally speaking, the percentage of quartzites decreases while the percentage of flint increases. Overall, debitage products in flint are relatively rare, whereas imported flint tools abandoned at the site are more frequent. These tools correspond to individual personal equipment brought to the site by human groups (personal gear, Binford, 1979). The presence of certain tools is only indicated by retouch and resharpening flakes, which demonstrate that they transited through the site (phantom tools, Cahen and Keeley, 1980 and Porraz, 2005). This is observed in particular for tools in the most distant raw materials, in flint from Tréviño and Urbasa for example, which were only identified by sidescraper sharpening flakes (Minet et al., submitted). This is particularly striking in Cjr-base, where the management strategies of tools in distant raw materials appear to be different from the ones in the other levels.

Variations in tool-type frequencies are also visible between the levels. The toolkit from level Cj is characterized by a low proportion of denticulates (16%), and a majority of sidescrapers (55%). The other tools consist of flakes with partial and/or marginal retouch, which are not characteristic of any particular tool type.

Sidescrapers are still predominant in level Cjgr, but there is also a clear increase in denticulates in this level. This is visible in the Cjgr toolkit in both zones (square 3F and in the test pit 6D/E–7E). The presence of cleavers on ophite flakes was also only evidenced in this level. These artefacts are an important component of Cjgr as they are absent from the surrounding levels. This provides an additional argument to refute the hypothesis that they result from the simple persistence of tool types from the Acheulean in the different regional Middle Palaeolithic industries (Cabrera Valdès, 1983, Cabrera Valdès, 1984, Freeman, 1966 and Freeman, 1969–1970). Conversely, it shows that they must participate in the characterization of the industry from which they came, like any other element. A low percentage of Quina type sidescrapers is also present in Cjgr in the test pit zone, but they probably result from disturbances with the underlying level Cjr-base, where the quantity of these tools increases considerably.

Indeed, the toolkit from level Cjr-base can be clearly distinguished from the other levels. This is the only level in which flint is the dominant raw material for the toolkit (46%), whereas quartzite only represents 41%. Sidescrapers are clearly prevalent, while denticulates are rarer. Quina-type sidescrapers are only made on flint. On the other hand, some sidescrapers in quartzite also bear demi-Quina retouch. Equivalent proportions of flint and quartzite sidescrapers display simple scalar retouch. Tool types with Quina-type retouch are mainly transversal sidescrapers, followed by double convergent sidescrapers and limaces, while demi-Quina retouch is more frequently associated with simple lateral sidescrapers. In addition, two plano-convex bifacial scrapers were also identified (Fig. 10.6–7). Finally, five flakes from sidescraper resharpening with retouch on their distal edge were also observed. This type of economic behaviour is one of the characteristics of the Quina technocomplex defined in the North of Aquitaine (Bourguignon, 1997, Faivre, 2008 and Jaubert et al., 2001).

Finally, in level Cr, tools in quartzite (64%) are much more abundant than tools in flint (36%). Again, denticulates and notches are almost absent. Sidescrapers are the best-represented tools, mainly simple lateral types. Several sidescrapers with Quina-type retouch are also present (Fig. 10); however, they are concentrated in the upper part of the level (Fig. 4). Therefore, at this altitude, we cannot completely rule out the possibility that they may initially have been part of level Cjr-base.