Centre for Language and Communication Research
Cardiff University
wraya@cf.ac.uk

In previous work (Wray 1998, 2000) I have explored the idea that socially sophisticated hominid populations could have communicated in complex ways without grammar, by using a holistic protolanguage: a phonetically-articulated system of discrete, agrammatical messages. This paper tests that scenario, by considering the ways in which such a protolanguage would restrict the expressive scope of its speakers, and the effects that this could have on the nature of hominid life and on the timing of further evolution. I shall demonstrate that, specifically, a holistic system is incompatible with the coining of names for things and people, and with the extensive use of declarative statements. The absence of these two features of referential expression would inhibit information exchange and could explain the long period of technological and cultural stagnation between 1.4 million and 100,000 years BP (Mithen 1996:116).

It seems reasonable to look for a role for holistic processing in protolanguage, since, not only did protolanguage presumably develop out of a holistic communication system of some sort, but it also developed into a communication conglomerate which, besides words and grammar, continues to support holistic processing (Wray 1999, Wray & Perkins 2000). Recent research on the nature and extent of formulaicity in human language (for reviews see Wray 1999, forthcoming) has revealed it to be a common and essential feature of our every day communication. Formulaic language has a number of communicative functions, of which almost all directly correspond with the manipulative socio-interactional functions observed in ape communication (Reiss 1989). Our socio-interactional formulaic expressions are directed towards the physical, emotional and perceptual manipulation of the hearer: commands, requests and warnings can all be expressed using fixed phrases; politeness is associated with fixed forms; we use in-phrases, chants and songs, etc. to indicate our group identity and so on (see Wray 1998, 2000, forthcoming, Wray & Perkins 2000 for a full exploration of these and the other functions of formulaic language). A significant advantage to us of using such fixed forms for common interactional functions, rather than constructing a novel sequence to express the same idea, appears to be the ease with which they can be recognised and decoded by hearers who share our language variety. The easier the decoding, the greater the likelihood of the hearer reacting in the desired way.

In this model, protolanguage messages are semantically complex and agrammatical. They are holistic, which means that a complete message is uniquely associated with an arbitrary form, not made out of smaller recombinable units of meaning. For instance, we might say that tebima means give that to her and mutapi means give that to me. Note that "there is no phonological similarity between sequences with similar meanings, because they are holistic. There is no part of tebima that means give or her. Simply, the whole thing means the whole thing" (Wray 2000).

Much has been written over the years about the power of naming, and it has seemed logical to assume that the forerunners of language included amongst their attributes, from the earliest stages, names for things and people. Hypothetical protolanguages are characteristically depicted as heavy on referential items when still lacking any means for expressing the complex relationships between them. But in a holistic protolanguage there is no place for naming, and the use of pronouns in the glosses of tebima and mutapi above is significant. If a holistic system like this tried to be too precise, such as referring to each member of the group and to each common object individually, it would soon run into difficulties. Without recombinable constituents, you would need an entirely new sound-string for each possible utterance about each person and thing – different ones for give the stick to Mary, give the stick to Joyce, give the stone to Mary and so on. Although highly precise in meaning, each individual string would have such a low functional load as to be rarely heard and used. This would make it difficult to remember, and once forgotten it could not be reconstructed by rule since there are no rules or individual words. Furthermore, this level of precision would generate so many strings that differentiating them would require an excessively large pool of phonetic variants in lengthy combinations.

These difficulties are resolved by using the same string for many speech events. By grouping together, say, all the female persons into one set, a single string give that to her can stand for give that to Mary, ... to Joyce, ... to Brenda and all the rest. Even now, the inventory of such strings will still be large, and the need to differentiate them could explain why we possess today a capacity for making articulatory distinctions far in excess of what we need for our modern analytic language of small, recombinable units (Wray 1998, 2000). There is a price to pay, however, for not being able to express the specific identity of individuals or objects. If you say give that to her, how will the hearer know to what, or whom, you are referring? For the purposes of disambiguation, such messages must be supported by indicative gesture, eye-gaze, or whatever. In short, although the speaker and the hearers all perfectly well know the difference between Mary, Joyce, Brenda and the other females, and between stones, sticks, meat, leaves and other portable objects, they are handicapped in their ability to express linguistically these fine distinctions by the limitations in the expressive potential of the language. The shortfall is made up using holistic paralinguistic strategies which we still put to the same purposes today.

Hot on the heels of having no names comes the effective absence of declaratives in all but one limited context. Even though there is no cognitive restriction on their wider use, there is a linguistic one. Because new items in the language can neither be created nor decoded by rule, the cost of introducing them is high, and there must be a good reason for bothering. Manipulative messages such as commands and threats offer a practical pay-off, and also are relatively easy to understand using pragmatic cues or by observing the behaviour of speaker and hearer. In contrast, the function of declaratives is to change others’ knowledge and, for them to be useful, the speaker must be imparting some information that the hearer does not possess, and could not have deduced. If you cannot specifically name, the effectiveness of declaratives is drastically compromised. Since an utterance that could refer to any of several referents is incomprehensible unless the speaker can indicate who or what is meant, the referent must be present. Yet if the referent is present, any information that is observable need not be explicitly imparted. A general message such as look at her will be more economical and just as effective as separate messages for she is tall and she is limping. This leaves one condition, where the referent is present but the new information about him/her/it is hidden (e.g. she is sick; she is sad). The imparting of privileged information is a type of gossip (Dunbar 1996, 1998), oiling the wheels of social cohesion and adjusting the relative social status of speaker and hearer. It is closely allied with the other manipulative uses of language which are central to the continuity of holistic processing from primate to human. In the case of gossip only, then, just as Dunbar (1998:105) suggests, there is both the interactional motivation and the linguistic means for using declaratives.

This position offers a new perspective on Tomasello & Camaioni’s (1997) exploration of declaratives, which they, also, view as a special case in language evolution. Their stance differs from the one in this paper, since they consider that the communicative ability to express declaratives is held in check by an absence of the cognitive means, whereas I am proposing the reverse. A problem for their position is that declarative gestures are virtually absent in primates in the wild, but present in some apes taught a human-like language system. This undermines their claim that declaratives are beyond the cognitive capacity of apes. However, it is consistent with the cognitive and expressional functions that underlie the use of declaratives being dissociable, so that untaught apes, like the hypothesised protolanguage users, are prevented from making declarative gestures/utterances not because they lack an appreciation that others have separate mental representations, but because, lacking naming, they have little if any use for declarative gestures. This is what Gómez et al (1993: 419) suggest, though for different reasons: declaratives are a skill for which apes naturally have the potential, but normally no motivation. Meanwhile, apes that are taught to name gain both the means and the motivation for using declaratives.

The holistic protolanguage described above is a self-supporting, stable edifice, whose constraints would stifle its own further evolution, perhaps for thousands of generations: specific naming is unsustainable; without naming, declaratives have almost no purpose; without declaratives, information exchange is largely impeded; this minimises technological and cultural innovation, rendering naming unimportant. It is possible to imagine how the combination of strong social hierarchies, themselves very well served by holistic utterances, and this stranglehold on referential expression, could considerably postpone the breakthrough into full human language, relative to the independent evolution of our modern cognitive and intellectual abilities.

By postponing the linguistic expression of the ‘naming insight’ until hominids are on the very brink of anatomical and mental modernity, there is no need for us to postulate a gradual development of grammar simply in order to accommodate major neurological changes. Provided it is conceivable that our modern mental capacities arose independently of language, naming is unleashed into a powerful cognitive forum that can immediately exploit referentiality by creating argument structure out of the juxtaposition of a word and a holistic utterance, and by segmenting holistic utterances to ‘identify’ new words and structures post hoc (see Wray 1998, 2000). Importantly, the significance of naming in this context is not the coining of an arbitrary phonetic sequence with a consistent meaning, for that hurdle was overcome long ago, when the holistic messages were first coined. What is different about naming is the size of the meaning unit: it refers to a person or thing (and soon an idea, action, attribute, etc) rather than a proposition.

Taking on board all the provisos concomitant with the view that "ontogeny epitomises (although it does not recapitulate) phylogeny" (Studdert-Kennedy 1998:172), it may be noted that first language acquisition follows pretty well the same path at the one outlined above. First, infants use holistic noise and gesture for manipulative and social purposes. They then pick up contours and articulatory features of complete utterances, which they produce as holistic phonologically-governed message-carriers. When they gain the naming insight they become referential in gesture and word, and proceed to segment the holistic strings into smaller meaning units that can be juxtaposed to create grammatical relationships. However, they continue to employ a holistic processing strategy for various ends, including the gamut of socio-interactional functions associated with the physical, emotional and perceptual manipulation of others (Wray 1999, forthcoming). The significance of this similarity between ontogeny and phylogeny lies primarily in the superimposition of naming onto an existing holistic system, and the way in which referentiality and compositionality team up with formulaicity to maximise the advantages novel and routine expression (see Wray 1998, 2000, forthcoming; Wray & Perkins 2000).

This scenario offers several explanatory advantages. It accommodates a slow development of long-term memory storage and of our excessive articulatory prowess, under pressure from a greedy communicational system that takes no hostages (if you can’t distinguish messages, or remember them, there is no grammar to fall back on). It provides a long period of phonetically-articulated complex social interaction without a linguistic facility for information exchange, at a time compatible with the technological and cultural stagnation identified from the palaeontological record. It holds back the onset of grammar to a point where it can develop quickly to its present state, while never entirely usurping the ancient communicative roles of holistic messages, thus avoiding scenarios in which simple grammars are of questionable use to speakers with complex communicational needs. It convincingly mirrors the patterns of child language acquisition, particularly in depicting the holistic system as more primitive. And it provides a link from the holistic approach to communication used by primates, and presumably by our earliest ancestors, through to our use of formulaic language as a significant and necessary supplement to our grammatical system today.

 Conference site: http://www.infres.enst.fr/confs/evolang/