Interjections (ouch, oops, wow, yuk, eh, oh etc.) are a fairly broad class of items which have been variously defined. The questions of what and how they communicate, and whether or not they are part of language, have been addressed from two largely dichotomous viewpoints. Ameka 1992, Wierzbicka 1992, Wilkins 1992 treat interjections as encoding rich conceptual structures which are part of the semantics of natural language. Goffman 1981 regards them as falling outside the linguist’s domain, analysing them in terms of the socio-communicative roles they play. My aim in this paper is to assess the relative strengths and weaknesses of these two contrasting approaches, and to suggest a new analysis of interjections which preserves the insights of both. This analysis has implications for accounts of the evolution of language.

While Ameka et al. are agreed that since they have semantic structure, interjections are part of language, they do not agree on what exactly constitutes an interjection; Goffman too has his own criteria. There is, however, broad agreement from both camps that interjections satisfy the following two criteria: firstly, they are capable of constituting an utterance by themselves in a unique non-elliptical manner; secondly, they express a speaker’s mental or emotional state or attitude. There is also agreement that one can distinguish two types of interjections: primary interjections (ouch, oops etc.) – words that cannot be used in any other sense than as an interjection, and secondary interjections (shit, damn etc.) – words which have an independent semantic value but are often used as interjections. Primary interjections are the main focus of this paper.

For the conceptualists, interjections have " ‘real’ semantic (i.e. propositional / conceptual) content" (Wilkins 1992: 119). The kind of analysis they propose is exemplified below – (Wilkins’ proposed conceptual structure for wow (151) (1):

1. "wow!"

I have just now become aware of this something,

There are four main problems with this approach. Firstly, the communicative content of interjections is highly context-dependent, and an utterance of wow seems to communicate something altogether more vague than the elaborate conceptual structure in (1). Secondly, humans use a wide range of para- and non-linguistic behaviours to communicate: facial expressions, gestures, some intonation, for example. The conceptualist approach overlooks the fact that interjections share with these the property of being partly natural as well as partly coded. Thirdly, interjections do not contribute to the truth-conditions of sentences that contain them, and this is surprising if they encode fully-conceptual structures. Conceptual representations have logical properties (they can contradict or imply other conceptual representations); one would therefore expect the constituents of these representations to be truth-conditional. Fourthly, intuitions do not support the claim that interjections encode conceptual structure. Whilst one is happy to concede that the italicised expressions in (2) encode the same (or similar) concepts, one is less happy to say the same about those in (3ab), which do not appear to be synonymous in the same way:

(2) Be careful with that needle / hypodermic / syringe.

(3) a. I feel pain, what did you do that for?

b. Ouch, what did you do that for?

Furthermore, (4a) below intuitively involves a conceptual repetition not evident in (4b):

4. a. I feel pain, I feel pain.

b. Ouch, I feel pain.

Goffman defines interjections, or response cries, in functional terms. Some are entirely instinctive, natural reactions – brrrr, for example – the purpose of which is to restore some kind of physical equilibrium. Others are less instinctive; a person uttering oops on dropping something might do so because it has the effect of "downplaying import and hence implication…of…incompetence" (1981: 102). Goffman also regards swear words (imprecations) as response cries and introduces the notion of a continuum: "response cries such as eek! might be seen as peripheral to the linguist’s domain…but imprecations …are more germane, passing beyond semi-word segregates to the traditional material of linguistic analysis" (p.121). However, Goffman’s account fails to address the question of how interjections communicate what they do.

Over the last 30 years philosophers of language and linguists have broadly agreed that there are two distinct kinds of linguistically encoded information. The distinction has appeared in various forms – truth-conditional versus non-truth-conditional meaning, describing versus indicating – and is one we might exploit in developing a more adequate analysis of interjections. One version of this distinction is the one made in Relevance Theory (Sperber & Wilson 1986, 1995) between conceptual and procedural encoding. Relevance Theory is an inferential account of communication within a cognitive framework. The linguistically encoded content of an utterance (characterised as the output of a modularised language faculty (Chomsky 1995)) is seen as providing no more than a starting point for rich inferential comprehension processes guided by expectations of relevance, cognitively defined: linguistic communication is massively inferential. This model can be contrasted with a code model of communication, where communication is achieved simply by coding and decoding. The conceptualist approach is reminiscent of such a code model, in which inference plays a rather minimal role.

The idea behind the conceptual-procedural distinction is that while most words encode concepts, constituents of conceptual representations, the function of certain others is to guide the hearer’s search for the speaker’s intended meaning by constraining the inferential processes that construct or manipulate those conceptual representations. As a first step toward a new account I propose that interjections encode procedural information. Rather than encoding the specific conceptual structure in (1), wow might encode a procedure which points the hearer in the direction of attributing to the speaker an attitude of delight or astonishment, without this being encoded. This approach solves three of the four above problems with conceptualist approach. We are left with the problem of how best to characterise the fact that interjections share certain properties of para- or even non-linguistic behaviours: even if we can analyse interjections as encoding procedural information, they are surely not deserving of the same kind of linguistic status as other (non-interjectional) procedural expressions.

We may be able to capture the uncertain linguistic status of interjections (and also the heterogeneous nature of the class) by drawing on Goffman’s notion of a continuum. Thus we might propose that interjections are located at various points along a continuum between showing and telling, where showing is relatively natural behaviour and telling is properly linguistic. Consider an example: to show Jack that she is delighted with a gift, Lily allows him to see her natural reaction, a smile; to tell him, she might utter something like ‘it’s wonderful’. To utter an interjection such as wow is to communicate delight with a degree of encoding which takes it beyond mere showing, but falls short of telling. I further suggest that the procedural approach might be extended to some of the para- and non-linguistic behaviours discussed above: it follows, therefore, that interjections may encode procedural meaning whether or not they are properly linguistic.

Fundamental to the relevance theory model of communication is the ability of humans to attribute thoughts, intentions and beliefs to one another: the ability to entertain metarepresentations, that is, representations of other representations. An act of ostensive communication is achieved by a speaker providing evidence of her intention to inform the hearer of something. In any ostensive act there are two layers of information to be retrieved. The first layer is the information being ‘pointed out’; the second is the information that the first layer has been pointed out intentionally.

When Lily smiles, she provides direct evidence of the first layer of information she wishes to convey: she shows Jack that she is delighted with the gift. He can see that she is delighted without necessarily paying attention to her intention to inform him.Lily’s linguistic utterance constitutes a very different kind of meaning, however. Here, Jack’s recognition of Lily’s intention to inform him is of crucial importance. Since the relationship between her utterance and the fact that she is delighted is arbitrary – it is a coded signal – her utterance provides only indirect evidence of the first layer of information. Instead, all the evidence provided by her utterance bears directly on her intention to inform Jack. Seen in this way, the distinction between showing and telling reflects Grice’s (1982) distinction between what he termed ‘natural’ and ‘non-natural’ meaning, one feature of which was that increasingly complex metarepresentations might be entertained in a communicative act.

Metarepresentation has a psychological function independent of communication, that of predicting the behaviour of con-specifics (Byrne and Whiten 1988), perhaps in such a way as to better integrate social groups (Dunbar 1998). Since the metarepresentational ability could have evolved independently of communication, one might argue that it preceded, rather than followed, the evolution of a language faculty (Sperber (forthcoming)); having some metarepresentational ability, after all, makes inferential communication possible in the absence of language. To individuals already deeply involved in inferential communication a language faculty may have proved adaptive by making the input to evolving inferential abilities more explicit. In the absence of language, however, communication would have involved the use of other behaviours such as gesture, facial-signalling or mime (Donald 1998). On this approach, the evolutionary significance of the showing/telling continuum is clear: showing came before telling or saying that. The showing/telling continuum may represent an evolutionary progression from natural communicative behaviour to language proper.

There is evidence to support this claim. Neurological evidence suggests that interjections are associated with phylogenetically ancient sub-cortical circuitry, as opposed to more recent cortical structures implicated in the production of language. The relative phylogenetic histories of the two systems may differ in a manner analogous to that proposed for human vision (Ramachandran and Blakeslee 1998), where as well as the evolutionarily more recent ‘visual pathway’ to the visual cortex, responsible for the lion’s share of visual processes, there exists a phylogenetically primitive pathway to the superior colliculus, which mediates unconscious orienting behaviour. This, together with the fact that interjections such as ouch remain within the repertoire of certain serious aphasics (Goodglass 1993), demonstrates a clear dissociation between at least some interjections and language proper. The degree to which the showing/telling continuum might be reflected in the neural structures that mediate communicative events needs further investigation, and may be illuminated by increasingly revealing techniques of neural-imaging.

Adopting a broader perspective, this paper supports the view of Jackendoff (1999) who suggests that some sort of "middle ground" needs to be defined between Chomskyans, who, despite advocating an innate language faculty, have in the past tended to "devalue evolutionary argumentation" (p.272), and evolutionists, who have been inclined to deny the possibility of an innate language capacity in the absence of any evolutionary justification. Jackendoff also suggests that interjections might represent "fossils of the one word stage of language evolution" (p. 273). The proposals here develop this suggestion.

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