A global profile of each assemblage is provided so that we may more clearly observe morpho-technical variation in the Caune de l’Arago industries, thus contributing to defining the specificity of each occupation. The main criteria discussed include: raw material selection patterns and use, technological features, and typological factors. Some of the archaeostratigraphic levels presenting analogous techno-typological features have been grouped together in order to simplify this comparative study (Table 2). This analysis synthesizes data obtained from nearly 100,000 industries exhumed from 14 stratigraphically distinct occupation levels (Barsky, 2001, Barsky and de Lumley, 2010 and de Lumley and Barsky, 2004; Table 2, Fig. 2). Levels were attributed to each stone artefact based on observations recorded in field notes during excavations (Barsky, 2001). Fragments, mostly in quartz, represent more than half of the material considered in this study (> 60,000 pcs) and have yet to be quantified into a global inventory per level. However, they constitute an essential component of the Caune de l’Arago assemblages and their study has largely contributed to our knowledge about core morphology and knapping strategies at this site. Our work on the fragments has also allowed us to monitor the sheer volume of raw materials brought into the cave. For the purpose of this study, each fragment was individually measured and described; recorded features include: raw material, size, corticality and formal morphology. Raw materials were determined for each piece using a referential elaborated during petrographical studies (Grégoire et al., 2006 and Wilson, 1988).

Looking beyond the massive quantity of fragments, the overall representation of other types of stone waste (Table 2, Fig. 2) shows that flakes, cores and small retouched tools are well represented in all levels. Most levels contain a macro toolkit with more or less intensively modified/shaped pebbles. All stages of core reduction are represented in each ensemble, attesting to in situ knapping, retouched tool shaping and pebble-tool manufacture. Cortical flakes are, however, scarce among all rock types excepting limestone, suggesting that cores were often initiated outside of the cavity while pebble-tools (almost always in limestone) were shaped inside the cave. An increase in core frequency is noted towards the top of the sequence and may be explained by the more extensive use of small sized quartz pebbles in these levels (Table 2). The pebble-tool assemblage is largely dominated by choppers of various morphologies while chopping-tools are very scarce. Retouched tools are most frequent in the upper levels (C, B, A). However, this result may be biased; assemblages from the upper part of the sequence are incomplete because the infill was truncated and eroded after the collapse of the cave's roof (notably levels B and A, de Lumley et al., 1984).

Handaxes (Fig. 3 and Fig. 4) and (especially) cleavers are poorly represented overall. They are absent from Unit II (total number of pieces studied from Unit II = 7740, Barsky, 2007), except for a single, poorly standardized piece in hornfel from level J whose morpho-technical characteristics are closer to a core than to an intentionally configured tool (Barsky, 2007). While most of the handaxes were shaped from pebbles, particularly those in hornfel, some of the pieces were configured on flakes (notably in quartzose sandstone). The extensive use of river pebbles may be one explanation for why the production of large flakes intended for the realization of LCT's is not a characteristic of the Caune de l’Arago industries (Sharon, 2009). Some of the handaxes are of very small dimensions. Handaxes show highest frequency in the older, P levels assemblages. Cleavers are extremely rare and poorly standardized throughout the entire sequence (6 pieces: 1 from level P, 3 from level G, 1 from level F and 1 from level E). As in other western European Mode 2 sites of similar chronology (Notarchirico, 0,6 Ma, Italy, Piperno, 1999), handaxes are absent from some levels (Unit II levels and level K in Unit I).

The bulk of our study (Barsky, 2001) focuses on flake morphology, taking into account: extension/localisation of cortical residue on dorsal surfaces, type of striking platform, orientation of removal negatives, average dimensions, and striking angles. Different aspects of the other components of the industry, notably cores and pebble-tools, have been taken into consideration to contextualize our detailed flake analysis. The quantification of the various aspects of the flakes is therefore considered in relation to the rest of the industry and in accordance to the different raw materials used. Archaeometrical comparisons between brute and retouched supports help to define tool support selection criteria. The methodology defined for core classification is outlined below, with special attention given to the technical criteria elaborated during our study.

The Caune de l’Arago industries show high petrographical diversity (in order of abundance): vein and translucent quartz, flint (or siliceous sedimentary: including flint, lydian and jasper), quartzite, quartzose sandstone, sandstone, limestone, hornfels, and lava. While most of the rocks were gathered by hominins as river cobbles of various sizes, most siliceous sedimentary rocks (heretofore flint) were collected from outcrops considerable distances away. Poor quality flint was nevertheless accessible from a handful of local or semi-local sources in the form of small pebbles or nodules (Verdouble alluvials and nearby outcrops). Flint more apt for controlled knapping was accessible in the form of large plates or blocks from outcrops at least 30 km away (Grégoire et al., 2006 and Wilson, 1988). As a rule, the rocks available close-by were poorly adapted for controlled flake extraction (quartz, sandstone, quartzose sandstone, limestone) while more suitable materials (jasper, flint, quartzite, blue translucent quartz) were accessible further away from the site (15 to 30 km). Although there is little change in the frequency of these different rock types throughout the stratigraphy, detailed analysis of each of the 106 sub-types composing the nine main petrographical groups represented does provide interesting relative frequency results; perhaps reflecting raw material preferences over time (Barsky and Grégoire, 2001). Amongst the local rock types, vein quartz was the preferred raw material in all levels. This may be because it provided more resistant cutting edges compared to other, easily accessible materials (∼90% of the level G industry). While in the present day Verdouble river bed vein quartz is relatively scarce, it is found in greater abundance in alluvial terraces believed contemporary to the cave's habitation. Both vein and translucent quartz pebbles were collected in the form of small (fist-sized) to medium sized pebbles (max. L = ∼30 cm). Our study shows that each raw material played a specific role in the assemblages (Fig. 5). Interestingly, the raw material/tool type relationship was maintained over time: limestone was invariably preferred for heavy-duty tools and hammerstones, fine-grained siliceous rocks (flint, quartzite) were chosen for more complex and exhaustive knapping sequences (discoid) and for the retouched tools. Handaxes, mostly in hornfel, show the greatest raw material variability.

Varying raw material quality affected core and flake morphology, dictating technological constraints throughout the different phases of stone reduction. The dominance of vein quartz in particular helps to explain why some features of the Caune de l’Arago assemblage appear unchanged over time. For while it is true that some of the industries’ characteristics do remain constant throughout the stratigraphy, there are essential differences that can only be brought to light when each level is considered independently.

The relationship between raw materials and core reduction strategies, for example, always present the following characteristics: •

tested or partially exploited cores tend to be in limestone or sandstone. Scars on these larger sized cores attest to their auxiliary use as percussion tools;

The wide technological diversity and long stratigraphical sequence at the Caune de l’Arago presents the opportunity to effectuate significant intrasite comparisons. We examine subtle variations in raw materials, technology and typology from one level to the next in order to appreciate each ensemble within the larger framework of Mode 2 in southern Europe.

While some factors remain unchanged throughout the Caune de l’Arago's thick stratigraphical sequence, others do change according to the different occupation levels. We do not advocate the idea of a progressive evolution in the stratigraphy, since this hypothesis must be tempered by considerations regarding external impact factors such as climate change, hunting practices and raw material preferences. Characteristics that remain unchanged throughout the stratigraphy include the following ones: •

raw material sources and the relationship between rock and tool types;

Amongst the factors that do show some variation within the sequence, we note the following ones: •

an overall decrease in product size throughout Unit III with an almost microlithic industry in levels D through A. Assemblages at the base of the sequence (P levels) also show small average dimensions;

Handaxes are symmetrical and finely worked in the P levels, they are absent from Unit II and then become an irregular occurrence in the rest of the sequence. Pointed and converging-edge choppers appear at the base of Unit III. While retouched tools show little variability (Byrne, 2004), the associations between different tool types and overall tool morphology, rather than the tool types themselves, may help to distinguish transformative characteristics: scraper versus notched tool proportions, frequency of multiple tools or converging-edge shaped tools, presence and morphology of handaxes, toolkit diversity, tool standardization and quality of retouch, are all characteristics which show diachronic developments over time. Concerning flake production, discoidal flaking was performed using alternate surface technology in Unit I while discoidal core surfaces become asymmetrical from Unit III onwards with separate preparation (cone shaped) and extraction (slightly convex) surfaces. From level G, quartz shattering (versus controlled bipolar on an anvil in P levels) is the most common knapping method used and voluntary, massive production of fragments is manifest. Clactonian-type knapping methods, rare in P levels, dominate in level G, and are less common towards the top of Unit III. Globular cores, rare in Units I and II, are typical in level F and progressively more common towards the top of the sequence.

Levels defined as short-term stays (Table 1) have sparse accumulations of lithics, species-specific hunting and a relative frequency of carnivore remains. The industries share the following characteristics: initial stone reduction stages were executed off-site while products were finished inside the cave. Fine quality rocks are more common, even if they originate from sources further away from the cave. Knapping strategies are complex (multiplatform or discoidal), probably aimed at obtaining maximal exploitation.

In levels interpreted as seasonal stays (Table 1) where a single species was preferentially hunted, rocks were mostly collected from the immediate vicinity of the cave, although there are some imports from far away sources. There are numerous retouched tools, often shaped from fine quality rocks. The industries include heavy-duty tools, with whole and broken pebbles used as hammerstones. In spite of scraper dominance, retouched tools show fine workmanship and diversity. Knapping was carried out by multidirectional (exhaustive) methods, mainly discoidal, most often applied to finer rocks and Clactonian methods for knapping rocks poorly adapted for controlled flake extraction. Quartz was largely exploited by the bipolar on an anvil technique, but was also used for other, direct hammer core reduction methods.

In the G level, the faunal assemblage seems to translate longer-term stays since faunal remains cover different seasons of the year. Many different kinds of animals were hunted and the artefact accumulation is very dense (Barsky et al., 2005, Grégoire et al., 2008, de Lumley and Barsky, 2004 and de Lumley et al., 2004). Looking at the industries, raw materials are mostly local but some rocks were brought from distant sources. These, finer quality rocks were economically used (exhaustive knapping) and even small flint nodules and quartzite fragments were crushed on anvils to maximize their reduction. The industries comprise numerous non-modified elements such as: stones, anvils, whole and broken pebbles and numerous pebble-tools. Retouched tools are abundant and highly diversified, often with several tools on a single support (multiple tools). The overall aspect of the toolkit shows high diversity; there are finely worked tools as well as expedient or poorly finished ones. Knapping techniques were also very diverse, in spite of prolific quartz crushing. Poor quality materials were reduced by Clactonian rather than multiplatform methods and sequences were non-exhaustive. This partially explains the large average size index for these industries. The lithic assemblages seem to reflect that a larger array of activities was performed at the site. This hypothesis has recently been further supported by a study showing variable dental wear patterns on game animal teeth (Rivals et al., 2009).

We conclude that, at the Caune de l’Arago, a more “specialized” lithic assemblage is characteristic of shorter-term stays with species-specific hunting and that, inversely, a more diversified toolkit corresponds to longer-term stays with a wider array of hunting practices.

In spite of the importance attributed to the appearance of handaxes and cleavers, these tools cannot constitute the sole indicators of a single cultural complex. Although they are emblematic of Mode 2, handaxes and cleavers exist in stone-tool assemblages in different regions of the world from about 1.7 to 1.5 my (Africa, India) up to the end of the Middle Palaeolithic, and cannot therefore serve as viable chrono-cultural markers. Outside Africa, the lower chronological limit for the appearance of Mode 2 varies from 1.6–1.4 my in the Near East (Ubeidiya, Bar-Yosef and Belmaker, 2011, Bar-Yosef and Goren-Inbar, 1993, Martínez-Navarro et al., 2012 and Tchernov, 1992) to around 900,000 years in China (Lanxian, in Shaanxi Province and Yunxian, in Hubei Province; de Lumley and Tianyuan, 2008) and in South China's Bose Basin (Yamei et al., 2000). At the Caune de l’Arago, the P levels are a rare example of how innovative Mode 2 techno-typological features and associated behaviours took root in western Europe around 0,7–0.6 my (P levels of the Caune de l’Arago, Barsky and de Lumley, 2005 and Barsky and de Lumley, 2010; Middle Loire River Basin, Despriée et al., 2009; Notarchirico, Venosa; Piperno, 1999). Globally, the production of Mode 2 industries marks a new phase in human development and signals the appearance of a greater abundance of sites (after ∼500,000 years, Dennell, 2003). Work is still needed to understand whether the appearance of Mode 2 assemblages in different parts of the globe reflects a learned or imported phenomenon (Sharon, 2009).

Presently, archaeological evidence in western Europe dating to between 0.7 and 0.5 my is scarce and indicates that the appearance of Mode 2 technology occurred there around this time (Barsky and de Lumley, 2010, Despriée et al., 2009, Despriée et al., 2010 and Piperno, 1999). Some sites dating to within this timeframe lack handaxes and cleavers or show only alternate presence of these tool types: Isernia La Pineta, Caune de l’Arago, Vértesszölös, Bilzingsleben (Barsky and de Lumley, 2005, Barsky and de Lumley, 2010, Coltorti et al., 2005, Mania, 1990, Peretto, 1994 and Svoboda, 1987). At the same time, these sites present a range of other characteristics linking them to Mode 2, such as: enlarged raw material diversity and greater acquisition territory, diversification and standardisation of small and large retouched toolkits, use of discoidal and/or hierarchical flaking technology. In spite of the chronological and geographical gap separating Western Europe from the oldest Mode 2 in other areas of the globe: ca. 1.7 my in Africa, ca. 1.5 my in India, ca. 1.6 my in the Levant and ca. 1 my in North Africa (Bar-Yosef and Goren-Inbar, 1993, Beyene et al., 2013, Lepre et al., 2011, Pappu et al., 2011 and Raynal et al., 2001), some or all of the techno-typological changes defining this techno-complex could have occurred in Western Europe within the context of demographical continuity, since the presence of hominins from at least 1.4–0.8 My is now well attested at a growing number of sites – in Spain (Orce, Atapuerca, Vallparadís), France (Le Vallonnet, Pont-de-Lavaud) and Italy (Pirro Nord, Ca’ Belvedere di Montepoggiolo) and others (Berger et al., 2008, Bermúdez de Castro et al., 2010, Bermúdez de Castro et al., 2011, Carbonell et al., 1995a, Carbonell et al., 1995b, Carbonell et al., 2001, Carbonell et al., 2005, Carbonell et al., 2008, Despriée et al., 2006, Duval et al., 2011a, Duval et al., 2011b, Duval et al., 2012, Falguères et al., 1999, Falguères et al., 2001, Garcia et al., 2011, Garcia et al., 2012, de Lumley et al., 1988, Ollé et al., 2013, Parés et al., 2006, Rodríguez et al., 2011, Rosas et al., 2001 and Toro-Moyano et al., 2013).

A important feature shared by a number of early Mode 2 African and South Asian sites is the practice of large flake production for LCT's (Leakey, 1971, Pappu et al., 2011 and Raynal et al., 2001). LCT's are also documented at some sites in the Levant, with chronologies varying between 1.6–1.4 and 0.78 my (Bar-Yosef and Goren-Inbar, 1993, Goren-Inbar, 1998, Goren-Inbar et al., 2000, Martínez-Navarro et al., 2012 and Verosub et al., 1998), while others represent quite different knapping technologies (Bizat Ruhama, 1.2 Ma, Levant, Ronen et al., 1998 and Zaidner et al., 2010). Yet this feature is not shared by most western European Mode 2 sites; except in the Iberian Peninsula where it appears relatively late and where some knapping methods and morpho-types resemble those from some older North African sites (OIS 9, Sharon, 2011). This might point towards cultural transmission or replacement by North African populations moving into western Europe; perhaps via the Straits of Gibraltar. Although the origins for western European Mode 2 remain uncertain (local or regional evolution, cultural transmission or renewed colonization), demographic growth is clearly perceptible in the archaeological record after around 0.5 my, underlining the adaptive advantages of new techno-functional behaviours and justifying their widespread implementation. Presently, the debate about whether convergence or cultural diffusion (or both) were agents for technical transmission remains open.

Several Middle Pleistocene assemblages in Italy appear analogous to the Caune de l’Arago Unit II industries, such as those from Monte Conerno, Layer L (Bartolomei et al., 1966 and Peretto and Scarpanti, 1984), whose typological composition is similar: Levallois is absent and most cores are discoidal or polyhedral (Palma Di Cesnola, 1996). At Castro de’ Volsci (Biddittu, 1974, Biddittu, 1984 and Segre et al., 1984), industries present similar characteristics but also include symmetrical handaxes. The Isernia la Pineta assemblage (Crovetto et al., 1994, Peretto, 1994 and Rufo et al., 2009) contributes to defining Mode 2 diversity since it is devoid of handaxes: its unique morphology is partially attributed to petrographical features (tabular flint and limestone).

In France, the industries from the Middle Loire Basin (Despriée et al., 2006 and Despriée et al., 2009), the lower levels (H, I and K) of Aldène Cave (Cesseras, Hérault; Barral and Simone, 1976), la Baume Bonne (Quinson, de Lumley, 1969, de Lumley, 1976a and de Lumley, 1976b) and Orgnac 3 (Combier, 1967 and Moncel, 1999) show similarities in their structural development to the Caune de l’Arago Unit III assemblages, with some elements showing a progression towards Mode 3 alongside the presence of residual types (handaxes, choppers, Tayac and Quinson points).

In Spain, The Atapuerca Middle Pleistocene sites of Galería and Gran Dolina show a similar evolutionary trend in sites ranging from around 500,000–300,000 years old, while technology reflects occupational features (Ollé et al., 2013). At Dolina's TD10.2 where hunting strategies were focused on bovids, the lithic assemblage displays more specialized features (intensified use of chert). In the artefact-dense TD10.1 level, where activities appear more diversified and complex, raw materials are also more varied (Ollé et al., 2013). This accumulation has been interpreted to result from high intensity occupations (probably base camps) reflecting a wide diversity of on-site tasks (Carbonell et al., 2001, Márquez et al., 2001, Rodríguez, 2004, Rosell et al., 2011 and Terradillos, 2010). As with the Caune de l’Arago G levels, the TD10-1 accumulation seems to correspond to cyclical occupations with a somewhat palimpsest structure, where more divers toolkits reflect a wider range of on-site activities. The inverse trend of more finely worked handaxes in older levels and more crudely worked pieces in more recent levels is likewise observed at Dolina (Ollé et al., 2013).

The relative abundance of cleavers at Spanish sites such as Galería constitutes a significant difference not only with the Mode 2 assemblages of the Caune de l’Arago, but also with western Europe in general. Some authors have suggested a technological relationship and, by extension, probable migrations over the Gibraltar Straits at different times during the Pleistocene (Sharon, 2011). At Galería as at the Caune de l’Arago, small retouched tools are well developed although at the latter site their production occurred off-site. Differences in the technological choices (dominant centripetal at Dolina and Galería) may be attributed – at least in part – to the overall dominance of quartz at the Caune de l’Arago. Also, a regional-specific cultural shift from Mode 2 to Mode 3 is perceived in both the Dolina and the Caune de l’Arago major stratigraphical sequences. Finally, we note that there are strong techno-morphological similarities between the single quartzite handaxe from la Sima de los Huesos (Carbonell and Mosquera, 2006) and two quartzite handaxes from the Caune de l’Arago P levels (Fig. 3 and Fig. 4).

At the chronologically and geographically related sites of Cau del Duc and La Selva (Catalunya), differences in the industries are also attributed to site function variability (Carbonell, 1985). Elsewhere in Spain, in open-air sites near Madrid and in the Duero, Tage and Guadiana river basins, assemblages are dominated by flakes and include pebble-tools and handaxes shaped by direct hammer technology. Once again, unlike in France and Italy, these Spanish sites generally include more cleavers. The assemblage from Pinedo (Toledo) has been interpreted as representative of an archaic phase of the Acheulian (Santonja and Perez-Gonzáles, 1996), with pebble-tools, trihedrals and handaxes showing a low degree of symmetry. Discoidal flaking is prominent. Retouched tools are mostly scrapers with denticulates. Torralba and Ambrona (central Meseta; Gonzàles-Echegaray and Freeman, 1998, Howell, 1962, Howell, 1965 and Santonja and Villa, 2006), correlated to OIS 12 have also yielded Middle Pleistocene faunal and stone assemblages. At Torralba tools were made on pebbles from distant sources. Retouched tools from these sites and from the nearby Aridos I and II sites include numerous scrapers, as well as notches, denticulates and some points. Handaxes and cleavers are present and knapping is mainly discoidal. All three sites evolve towards a more standardized toolkit and Levallois technology towards the top of their sequences (Freeman, 1991, Gonzàles-Echegaray and Freeman, 1998 and Santonja and Villa, 2006).

During the Middle Pleistocene the typology and technology of stone industries develops and diversifies. Although this diversity certainly results from multiple factors, the transition from Mode 2 to Mode 3 industries in the Mediterranean basin does seem to follow some specific typo-technological stages, indicating a more or less coherent chrono-cultural context. The Caune de l’Arago industries may be placed within this complex and non-linear framework. Their richness and the extensive chronological period they cover make them a reference for several developmental stages of the Mediterranean Mode 2, while they express the typo-technological diversity of Lower Palaeolithic industries within a variety of paleoenvironmental conditions. The industries from the base of Unit I are amongst the earliest western European handaxe assemblages; the Unit II industries demonstrate that Mode 2 assemblages may not include handaxes and the Unit III assemblages show a regional progression towards a more standardized, Mode 3 type ensemble while conserving remnant Mode 2 features.