1.1. Language, cognition, and Palaeolithic stone tools
Modern humans are the only known species possessing language. Furthermore, in normal circumstances, all humans acquire at least one language before a certain developmental phase (cf. Levelt 2018; Pagel 2017). Thus, it can be hypothesized that language evolved no later than human speciation but not before the split from the last common ancestor of humans and chimpanzees. Language encompasses both the functioning of general bodily systems (breathing and digestion, musculoskeletal structures, hearing, cognition, etc.) and seemingly specific phenomena spanning from phonetics and phonology to morphology and syntax. Because language is such a multifaceted capacity, it seems plausible to hypothesize that language did not evolve in one “package”, but that specific language-related capacities appeared in the hominin lineage (or sooner) at different phylogenetic stages via different sets of mutations and evolutionary processes (which may have nevertheless been somehow intertwined; cf. Arbib 2016a, 2016b; Bickerton 1990: § 5–7, 2007; Botha 2020; Casielles & Progovac 2012; Collier et al. 2014; Dediu & Levinson 2013, 2018; Everett 2016: § 9–11; Hauser 1921: 113–122; Gabrić 2021a, b, c, d, e, 2022; Jackendoff 1999; Jackendoff & Wittenberg 2014; Krause et al. 2007; Lieberman 2015; McMahon & McMahon 2013: § 8–9; Michlich 2018; Planer 2017; Progovac 2015, 2016; Révész 1946; but see Berwick et al. 2013, Berwick & Chomsky 2016, Chomsky 2002: 161–162, Chomsky et al. 2019, Matasović 2012: 55ff., Tattersall 2019 for different opinions). If language indeed did not evolve in one “package”, it is plausible that some linguistic capacity was extant at least in the last common ancestor of modern humans and Neanderthals.
Despite these hypotheses and the recently reawakened interest in the evolution of cognition and language, hypothesis-testing research in this domain remains scarce. One controversial approach has been to identify behavioral correlates of specific cognitive functions in the archaeological and palaeoanthropological records (Bednarik 2013; Chase & Dibble 1987; Conrad et al. 2009; Criado-Boado et al. 2019; d’Errico 2003; d’Errico et al. 1998, 2005; Davidson 2013, 2014; Frayer et al. 2020; Henshilwood & Marean 2003; Hoffmann et al. 2018; Lindly & Clark 1990; Majkić et al. 2018; Prévost et al. 2021; Radovčić et al. 2015; Riel-Salvatore et al. 2001; Rigaud et al. 2009; Roebroeks et al. 2012; Tuniz et al. 2012; Turk et al. 2001; Watts 2009; Zilhão et al. 2010; see Bell 1994, Botha 2009, 2010, 2012, Garofoli 2014, Garofoli & Iliopoulos 2019 for criticisms). A large part of the research focus in this domain has been placed on Palaeolithic-stone-tool-related behaviors and it has been hypothesized by many that there might have been a co-evolutionary link between specific aspects of stone tool manufacture on the one hand, and the evolution of language and cognition on the other (Bruner et al. 2018; Davidson & McGrew 2005; Gabrić et al. 2018, 2021a; Lycett & Eren 2018; Moore & Preston 2016; Muller et al. 2017; Nonaka et al. 2010; Normile 2012; Nowell & Davidson 2010; Overmann & Wynn 2018; Parravicini & Pievani 2019; Stout 2010, 2011; Stout & Chaminade 2012; Stout & Hecht 2015; Stout & Khreisheh 2015; Toth & Schick 2009; Uomini 2015; Vaesen 2012; see Bar-Yosef 2017, Coolidge et al. 2015, Coolidge & Wynn 2001, 2005, 2016, Overmann & Coolidge 2019, Pain 2021, Wynn & Coolidge 2011, 2016, Wynn & Gowlett 2017 for different opinions). In other words, cognitive functions underlying linguistic capacity in modern humans – sensorimotor and visuospatial processing, working memory, declarative and procedural memory, executive functioning, cognitive planning, learning, etc. (cf. Hamrick et al. 2018) – could have been evolutionarily fine-tuned in association with the emergence or maintenance of specific stone-tool-related behaviors (e.g., manufacture and its acquisition), thus paving the way for the evolution of modern-like language (Lotem et al. 2017; Kolodny & Edelman 2018). Theoretical discussions have thus far proposed several cognitive “common denominators” for stone-tool-related behaviors and language; namely, action observation and execution (including sequential and hierarchical action processing; Adornetti et al. 2018; Arbib 2011; Brozzoli et al. 2019; Everett 2016: 80ff.; Greenfield 1991; Mahaney 2014; Moore 2010; Osiurak et al. 2020; Ruck 2014; Schlanger 1996; Vandervert 2018, 2020; cf. Kroliczak et al. 2021), auditory working memory and attention (Putt & Wijeakumar 2018; Vandervert 2018, 2020), social learning (Everett 2016: § 3–4; Gärdenfors & Högberg 2017; Lotem et al. 2017; Nishiaki 2019; Tennie et al. 2017; cf. Akazawa et al. 2014; Baronchelli et al. 2012; Botha 2015; Corbey et al. 2016; Gamble et al. 2011; Gibson 2012; Gowlett et al. 2012; Hewlett 2021; Mufwene 2018; Nonaka et al. 2010; Spike 2017; Rossano 2017), visuospatial processing (Bruner et al. 2018; cf. Schween et al. 2018), executive functioning and planning (Adornetti 2014; Ambrose 2010; Barham & Everett 2020; Everett 2016: 96–97, 112–113), etc.
Notably, a number of studies have indicated both anatomical and functional overlaps in the processing of real actions on the one hand, and action language on the other (see Arbib 2005, 2006, 2012, 2015, 2017, Arbib et al. 2014, Barsalou 2008, Frak & Cohen 2021, Ghio & Tettamanti 2016, Hauk et al. 2016, Kemmerer 2015, Pulvermüller 2013, 2018, Pulvermüller & Fadiga 2010 for reviews; Aziz-Zadeh et al. 2006; Boeckx & Fujita 2014; De Beni et al. 2005; Desai et al. 2010; Dreyer & Pulvermüller 2018; Dreyer et al. 2020; Geld 2006; Gianelli et al. 2020; Glenberg & Kaschak 2002; Grisoni et al. 2016; Hauk et al. 2004, 2006; Heard et al. 2018; Mollo et al. 2016; Progovac et al. 2018; Pulvermüller et al. 2005; Rizzolatti & Craighero 2004; Tettamanti et al. 2005; van Dam & Desai 2016; Willems et al. 2010; cf. Brauer et al. 2013; Hurford 2012; Kemmerer 2014). In other words, processing of action language is partly subserved by the functioning of the same brain areas and the same cognitive mechanisms which are involved in real action observation and execution. Action language refers to words and constructions which denote actions (e.g., the English verbs run, kick, play, attack, swing, dance, etc.) or other aspects of action events (e.g., agents and patients: attacker–attackee). Because Palaeolithic stone toolmaking is essentially a goal-directed action, it is plausible to hypothesize that there should be a positive association between behaviors related to stone toolmaking and action language use in modern humans. If this is so, a possible, yet controversial, evolutionary hypothesis would be that specific cognitive aspects of real action processing in the context of Palaeolithic stone toolmaking were evolutionarily reused for (action) language evolution.
Most studies investigating the relationship between real action and action language processing have focused on semantically transitive events and constructions. Semantically transitive constructions denote events in which an agent performs an action on a patient, with the latter experiencing some kind of transformation. Some English-language examples of semantically transitive verbs include break (e.g., He broke the glass.), cut, kill, hit, strike, kick, move, push, eat, take, etc. On the other hand, syntactically transitive verbs are verbs that are coded in terms of a subject and direct object and they do not have to be semantically transitive (e.g., I see a tree., She feels love., They crossed the street., She ran one hour., etc.; All of the four examples express semantically intransitive events.) (Creissels 2016; Fotiadou & Vassiliadou 2017; Saeed 1997: § 3.6, 6.1–6.6; cf. Palmer et al. 2010: § 1; Petruck 1996; Wright 2002).
The processing of semantically transitive verbs has been, among other areas, associated with increased metabolic activity in the pars opercularis of Broca’s area (Kemmerer 2012; Tettamanti et al. 2005; cf. Chen et al. 2021). While neural and cognitive correlates of syntactic transitivity are not well understood, some studies which investigated the processing of more elaborate transitive coding frames (such as those involving dependent object clauses) have found increased metabolic activity and/or structural integrity of the left pars triangularis of Broca’s area (Tyler et al. 2011; cf. Casado et al. 2020; Maran et al. 2021; van Dam & Desai 2016). Some have also hypothesized that because the supposed roles of the pars opercularis include, among others, representing actions at the conceptual level and representing the sequential and hierarchical organization of action concepts [i.e., agent–action(–instrument, patient, etc.)], there might have been a phylogenetic stage in which pars opercularis, or perhaps pars triangularis, applied these conceptual, sequential, and hierarchical processes via schematization to other cognitive domains, including more elaborate actions as well as language (Christensen 2010; Fazio et al. 2009; Fiebach et al. 2006; Gabrić 2021c, e: § 3; Kemmerer 2012; Kunert et al. 2015; Ruck 2014; cf. Ardila et al. 2016; Cohn et al. 2017; Cohn & Paczynski 2013; Fadiga et al. 2009; Foundas et al. 1996; Ghio & Tettamanti 2016; Grodzinsky 2000; Hopkins et al. 2017; Hupfeld et al. 2017; Liuzzi et al. 2017; Novén et al. 2019; Tate et al. 2014). Presumably, striking a core is cognitively processed as a semantically transitive event because an agent (knapper) performs an action (striking with a hammer) on a patient (core) and so changes the patient (core reduction). It is nevertheless possible that some level of action schematization is needed for the acquisition and execution of Palaeolithic stone toolmaking in modern humans, given that some have argued that action schematization is essential for the execution of at least some forms of Palaeolithic stone toolmaking (Herzlinger et al. 2017; Moore 2010; Schlanger 1996; Shimelmitz & Kuhn 2018; Sumner 2011) and that Hecht et al. (2014) reported that a two-year-long acquisition of Palaeolithic stone toolmaking was associated with fractional anisotropy changes in the pars triangularis. Thus, depending on the level of schematization of the actions involved in the acquisition and execution of Palaeolithic stone toolmaking, modern humans might show a positive relationship between the behaviors related to Palaeolithic stone toolmaking on the one hand, and either semantic or syntactic transitive language use (or both) on the other.
1.2. Oldowan
Oldowan is the earliest well-defined stage of hominin stone technology, ranging from ~2.6 (Semaw 2000, 2006; Semaw et al. 1997, 2003) to ~1.4 (Schick & Toth 2006; Toth & Schick 2018) or ~1.26 Ma (Semaw et al. 2020). The putatively earliest stone tools from Lomekwi, Kenya, putatively displaying simpler knapping techniques, have been dated to ~3.3 Ma (Harmand et al. 2015; Hovers 2015), yet their status remains disputed (Domínguez-Rodrigo & Alcalá 2016; but see Harmand et al. 2019 for a response to some of the criticisms), while Toth & Schick (2018: 7) list the artifacts from Lomekwi as part of the “Oldowan Industrial Complex”. Oldowan stone tool assemblages are typically characterized by the predominance of flaked core tools, namely unifacial choppers and bifacial chopping tools, flakes and other debitage, and battered percussors (Schick & Toth 2006). The majority of the knapping was done through hard-hammer percussion, either direct or bipolar, but the use of both the anvil technique and throwing against a hard substrate have also been proposed (Schick & Toth 2006). While some choppers and chopping tools may have been used for carcass processing (bone splitting and pounding) and other tasks, many such objects were possibly only cores (Shea 2020; Toth 1985). Regardless of their functional status, the knapping of these core tools and other core forms produced sharp-edged flakes which had the potential for use in butchering (Toth 1985, 1987). Retouched flake tools also appear in Oldowan assemblages along with unretouched flakes, although they are not ubiquitous or numerous (Schick & Toth 2006). Oldowan is primarily associated with the early Homo members (H. habilis and H. erectus), yet the contemporaneity of Australopithecus garhi occupation and animal bones displaying traces of breaking and cutting at Bouri, Ethiopia (de Heinzelin et al. 1999), with the stone tools in Gona suggest the possibility that A. garhi produced Oldowan stone tools as well (Semaw 2000; Semaw et al. 1997, 2005; cf. Schick & Toth 2006; Toth & Schick 2018). Furthermore, marks on bones found at the Dikika site, Ethiopia, strongly suggest that this behavior can also be attributed to Australopithecus afarensis (McPherron et al. 2010) and it is also possible that Paranthropus was one of the Oldowan tool makers (Susman 1991). Toth and Schick (2018) argued that Oldowan findings are suggestive of the incorporation of stone tools as a critical adaptive component which presumably led to more complex subsistence strategies, social behavior, and communication, while Michlich (2018) proposed based on an extensive review that Australopithecus africanus was capable of indexical, iconic, and possibly symbolic communication (“symbolic” in the semiotic, and, by extension, linguistic, i.e., Peircean and Saussurean sense and not in the sense of “symbolic” or “modern” behavior which is axiomatically inferred from artifacts and behaviors such as pigment use, personal ornaments, funerary practices, etc.; cf. Botha 2020).
Oldowan knapping behaviors display a degree of variation through time and space, and there has been a heated debate on whether there have been technological, cognitive, or different leaps within the Oldowan or if the variations in Oldowan knapping behaviors can be explained by environmental factors. One of the earliest proposals that there were technological leaps within Oldowan in the Olduvai Gorge was put forward by Leakey (1971, 1975) who differentiated between the so-called Classic Oldowan and Developed Oldowan A (DOA), with the latter supposedly associated with lower rates of choppers and increased rates of proto-bifaces, spheroids, subspheroids, and light-duty tools compared to the former. Following a similar line of thinking, Ludwig (1999) argued that the increased use of chert and increased manufacture of quartz spheroids and subspheroids during the DOA reflected enhanced knowledge of knapping mechanics and raw material selection, while de Lumley et al. (2009) proposed that some Oldowan sites older than ~1.9 Ma such as Gona, Lokalalei, and Fejej should be regarded as “Pre-Oldowan” because they display, among other putatively distinctive features, markedly high proportions of simple core tools and only sporadic retouch.
However, other research has suggested that these phenomena are best explained by raw material availability. Lithic analyses of the Bed I and II assemblages from Olduvai Gorge have convincingly shown that the incorporation of chert for stone tool manufacture in the Bed II Oldowan (1.65–1.53 Ma), when chert became available, elicited a considerable decrease in quartz and lava knapping and a considerable increase in chert knapping compared to the Bed I Oldowan (1.87–1.75 Ma), which turned back to Bed I Oldowan levels after chert became unavailable again (Kimura 1997, 1999, 2002; de la Torre & Mora 2014; McHenry & de la Torre 2018). Also, retouch was only limitedly present in the quartz and non-existent in the lava assemblages but was highly present in the chert assemblages (cf. de la Torre & Mora 2018). Chert was available at several locations in the Olduvai Gorge only during specific periods when the palaeolake Olduvai would retreat due to fluctuations in geological phenomena, exposing chert beds on the surface (Kimura 2002; McHenry & de la Torre 2018; cf. Stiles et al. 1974). The higher rates of flaking and retouch in chert-associated periods most likely reflect the fact that chert’s fine-grained texture makes it more suitable for both flaking and retouch compared to quartz and lava (Kimura 2002; McHenry & de la Torre 2018; cf. Spott 2005), indicating that the observed changes in knapping behaviors should be attributed to chert availability, rather than anatomical, cognitive, cultural, or other adaptations in the hominin populations (cf. Proffitt 2018). Furthermore, intersite variations in chert knapping behaviors have been explained either by chert availability or by differences in the locations on which different stages of the core reduction process took effect (Kimura 2002; but see Braun et al. 2005 for the latter). Thus, it would appear that the hominin populations which had access to chert were somehow anatomically, cognitively, culturally, and/or differently ready to incorporate it into their knapping behavior repertoire and adjust the repertoire. Interestingly, McHenry & de la Torre (2018) write that these hominins were “aware of the better flaking qualities” of chert compared to quartz and lava. Additionally, the putatively novel manufacture of quartz spheroids and subspherioids had in fact already been documented at pre-Olduvai Oldowan sites and discrepancies in the rates of quartz spheroid and subspheroid manufacture across place and time have been explained by differences in raw material availability as well (Semaw et al. 2009; cf. Sahnouni 2002; Sahnouni et al. 1997; Sahnouni & de Heinzelin 1998; Willoughby 1985).
Still, some Oldowan sites paint a slightly more complex, yet enigmatic, picture. Lithic analyses of the assemblages at HWK EE, a late Oldowan site dated to ~1.7 Ma, have demonstrated that knapping behaviors were characterized by shorter reduction sequences, lower flaking productivity, and simpler knapping methods compared to other Oldowan sites: “HWK EE hominins seem to have been uninterested and/or unable to exploit [quartz, lava, and chert] cores intensively and, if we are to use modern standards of productivity and efficiency, both were generally very low.” (de la Torre & Mora 2018; cf. Arroyo & de la Torre 2018; de la Torre et al. 2018; McHenry & de la Torre 2018; Pante & de la Torre 2018). It is thus possible that there were hominin populations in the Olduvai Gorge which displayed lower levels of either capacity or preference for more extensive stone-toolmaking-related behaviors, independently of the raw material type.
1.3. Previous experimental studies
Experimental studies of the putative co-evolution between stone-tool-related behaviors and language are found few and far between, probably due to serious methodological limitations. Many issues are epistemological, including the fact that the cognition of extinct hominin species cannot be directly assessed and valid model organisms for these species are currently non-existent (cf. Breyl 2020; Michlich 2018: 7; Putt et al. 2017: 1; but see Dannemann et al. 2020; Trujillo et al. 2021). Fortunately, however, the small number of previous experimental studies provide a valuable, albeit only fragmentary picture of how Palaeolithic-stone-tool-related behaviors are associated with modern brain and cognitive functioning. These studies can be tentatively divided into (1) teaching mode studies and (2) neuroimaging studies.
Teaching mode studies have mainly been trying to determine which modes of teaching, i.e., modes of transmission of information about stone tool manufacture in a teaching-learning context, can facilitate the acquisition of different stone-tool-related behaviors. Morgan et al. (2015) found that the acquisition of Oldowan-like chert flaking was facilitated in the gestural and verbal teaching conditions compared to imitation. Similar results have been obtained by Lombao et al. (2017) who investigated the acquisition of the alternating method applied on commercial bricks. However, Cataldo et al. (2018) found that Oldowan-like chert flaking was significantly better in the gestural and verbal teaching conditions compared to the (rather unnatural) gesture-free verbal teaching condition, possibly indicating that it is the gestures rather than spoken language alone that have a facilitatory effect on Oldowan-like (chert) flaking acquisition. Interestingly, contrary to the expectation that the acquisition of later stone technologies should then also be facilitated by either gestural or verbal teaching, Ohnuma et al. (1997) found no significant differences between the verbal and nonverbal teaching conditions in the acquisition of Levallois flaking on siliceous shale, while Putt et al. (2014) found that the verbal condition was associated with significantly worse chert flake quality during Acheulean-like biface manufacture compared to imitation.
Hecht et al. (2014) conducted a longitudinal diffusion tensor imaging study in which subjects spent two years acquiring Oldowan-like flaking, Acheulean-like biface manufacture, and Levallois flaking. Fractional anisotropy changes were observed in the left supramarginal and ventral precentral gyri, and the right pars triangularis (part of Broca’s area), indicating that the acquisition of the three Palaeolithic behaviors is associated with structural remodeling of inferior frontoparietal areas. In three studies using different neurophysiological techniques, it was found that Oldowan-like flaking compared to baseline is associated with increased metabolic activity in the cerebellum, secondary motor areas, and parietal areas, but not in the prefrontal cortex, tentatively suggesting that Oldowan-like flaking and its acquisition are not particularly executively demanding (Putt et al. 2017; Stout et al. 2000; Stout & Chaminade 2007). On the other hand, the acquisition of Acheulean-like biface manufacture relative to Oldowan-like flaking was associated with increased metabolic activity in inferior prefrontal (including the right Broca’s area), frontal, and parietal areas (Stout et al. 2008; cf. Putt et al. 2017). Putt et al. (2017) suggested that the association between metabolic activity in Broca’s area and Acheulean biface manufacture acquisition is limited to verbal teaching contexts, calling into question the role of Broca’s area in previous studies of Palaeolithic-stone-tool-related behaviors (cf. Bourguignon et al. 2018). However, these interpretations have been criticized (Gabrić et al. 2018: 14, 2021a; Uomini 2017). Be that as it may, because Broca’s area has a role in, among others, conceptual and schematic body representation, sequential and hierarchical goal-directed action processing, and aspects of linguistic and speech processing, it has been hypothesized that Broca’s area might have been a point of convergence in the evolution of language and stone-tool-related behaviors (Ruck 2014; cf. Kemmerer 2012; Rosenzopf et al. 2020). Increased metabolic activity in the prefrontal cortex during Acheulean-like biface manufacture compared to Oldowan-like flaking was not observed in expert knappers (Stout et al. 2008; cf. Stout et al. 2011), suggesting it is associated with acquisition rather than pure execution.
Converging data suggest that Acheulean-like biface manufacture may be more cognitively demanding than Oldowan-like flaking, with Oldowan-like flaking being apparently largely reliant on sensorimotor processing. Nevertheless, the involvement of the cerebellum and parietal areas might suggest an association with other cognitive functions as well. It is important to note that different parts of the cerebellum are related to both linguistic and speech processing (Mariën et al. 2014), attention (Kellermann et al. 2012), emotional control (Adamaszek et al. 2017), etc., while the parietal lobe, although known to be a seat of various functions (including linguistic and action processing; Bzdok et al. 2016; Deschamps et al. 2014; Fagg & Arbib 1998; McDowell et al. 2018; cf. Numssen et al. 2021), is likely functionally the least understood lobe. Furthermore, both the parietal cortex (Bruner et al. 2010) and cerebellum (Kochiyama et al. 2018; cf. Miura et al. 2014) have witnessed significant changes in size during hominin evolution. It is also very important to note that it is epistemologically problematic to infer cognitive processes from neuroimaging data (Poldrack 2006), especially of such broadly demarcated brain areas as in these studies. In fact, these areas are all associated with performances on tasks tapping into various cognitive capacities, suggesting that a more direct method is needed to establish a preliminary association between stone-tool-related behaviors and cognition.
The only experimental study thus far to compare neuroimaging data during a stone-tool-related behavior and performance on a cognitive task was by Uomini & Meyer (2013) who reported from their functional transcranial Doppler ultrasonography study with expert knappers high correlations between the hemodynamic lateralization patterns during Acheulean-like biface manufacture and silent letter fluency (naming as many words starting with a given letter), indicating similar patterns in the functional lateralization of particular linguistic functions and Acheulean-like biface manufacture (cf. Jöris & Uomini 2019; Osuna-Mascaró et al. 2020; Steele & Uomini 2009; Uomini 2015; Uomini & Ruck 2018; Vigneau et al. 2011).
In the present study, the aims were to (1) investigate the role of verbal and nonverbal teaching-learning contexts on the early acquisition of Oldowan-like quartz flaking, and chert flaking and retouch, (2) investigate the neuropsychological correlates of the early acquisition of specific technological steps during Oldowan-like flaking and retouch, and (3) specifically investigate possible associations between action language and Palaeolithic stone toolmaking. We were motivated by (1) the prevalence of use of chert as raw material in experimental studies (compared to, e.g., quartz), (2) lack of studies directly investigating which cognitive functions are associated with Palaeolithic stone toolmaking in modern humans, (3) the lack of experimental studies on retouch, as all have thus far focused on flaking, and (4) lack of studies on specific steps of the production process, as all have thus far focused on assessments of end products leaving open the question whether the observed processes are attributable to flaking in general or to some of its specific aspects. In this study, naïve subjects acquired to produce quartz choppers and chert sidescrapers in either a verbal or nonverbal condition. To explore the relationship between the performances on the stone toolmaking tasks and cognition, the subjects also performed on a battery of neuropsychological tests, assessing visuospatial working memory, cognitive planning, cognitive flexibility, and specific aspects of linguistic processing.