Celso Alvarez-Cáccamo

University of A Corunha, Galiza

AAA98, Philadelphia, Dec. 2-6, 1998

Session "Language Matters in Anthropology: A Lexicon for the Millennium", organized by Alessandro Duranti

Draft - November 1998

Author's page:
Author's address:

[ pdf.gif (159 bytes) Paper in PDF (46Kb) ]

     In his 1952 keynote address to the joint Conference of Anthropologists and Linguists in Indiana, Roman Jakobson stated that "The fundamental problem in the study of discourse is that of the commonality or sharing of the code between sender and receiver on which the exchange of messages is based" (Jakobson 1953, 1984:20. (1) Almost fifty years later, we all find ourselves here realizing that the problem is unsolved.

     What notion of "code" for anthropology? If, in our desert of skepticism, we are still meandering toward the utopian but mentally inescapable territory of Unified Science -- if we still sustain the reasons that led Sapir, Jakobson or Bloomfield to believe that language was a sign of the human -- then a code for anthropology should be what it should also be for linguistics or for communication theory.

     I believe there are basically two options in our crossroads here. The first option is a straightforward one: a code is just any linguistic variety -- a language, a register, a dialect, or a style. If we opt for this view, then I should simply yield my turn to Asif Agha or to John Du Bois, so that they could tell us again what a "register" or a "grammar" are.

     The second interpretation is the following: a code is something that is not a linguistic variety, a language, a register, a dialect, or the like. That is, certain attributes of a given code may overlap with certain attributes of a speech variety; then again, they may not. And, if a code cannot be structurally identified in terms of a linguistic variety, then perhaps a code is something of a different nature, and perhaps it is not even a structural entity, but a correlational function. To introduce my discussion, let me just advance that a code is, in a general sense, a correlational function, a rule, a communicative procedure.

     So once we have agreed on what a code is not, let's try to get at what, specifically, it is, and, how it works. I hope we can thus better understand why I chose the plural "codes" for the title of this talk, not the singular "code". For, even though we may be able to characterize what a code is, in human interactions one single code never happens: we function with and through a number of simultaneously operating communicative codes (whereas we are able to function just through one language).

     First of all: Where are codes located? That one is easy: codes are located between the brain and society. Humans, like other species, are inherently multimedia. We are the multimedia device par excellance. Even while we talk or sign in human languages, it would be excessive to think that a given mental state could be represented just through one language. The brain, if we have one, possesses mental states. These states of the mind/brain are not linguistic, but biological, organic, electrochemical in nature. They produce a sort of tepid ooze which we call cognition. Cognition contains a map of reality. Thus reality is actually contained in an ooze that emerges from a physiological mental state. But we still have no means of poking at our mental states. So we are left with language to believe that language is all there is to the mental representation of reality. But, who knows?

     That's all I know about the brain. Now, about society. There is a growing body of evidence that points to the fact that language-in-use sometimes just doesn't work as it should in indexing who we are socially, and what we are in interaction for. Briefly, this research reveals displacements in the way languages are used to index identities and intentions. That is, one may signal identity A by doing what appears to be identity B through language B. If identities and intentions could be transparently coded through language alone, these displacements shouldn't happen -- but they do. Communication, including talk, kinesics, proxemics, and other signaling systems, works primarily indexically. We rarely state who we socially believe we are: we just do our social selves, by alluding to categories of ascription. Thus, our interlocutors have to do a great deal of reconstructive interpretation, or inference. And this interpretation sometimes fails. Why does it fail? That is, if languages really are codes, understood as strictly closed structural devices, why do they fail? Because they don't fail. Because languages are not the codes we use to organize communicative intentions. Communicative codes are previous to language. If interpretation fails, it's not because codes fail. For codes are not "designed" to fail or succeed. Codes are inherently lose, inherently interpretive, inherently inferential, and inference is by definition a lose procedure, just as the cognitive ooze contains lose intentions which result in utterances and other things.

     Thus, we are in the position to advance a preliminary definition: a code is a correlational mechanism that transduces something into something else. The codes I'm talking about, communicative codes, transduce intentions into utterances, and then utterances into interpretations. Codes are previous to and relatively independent from languages and from other signaling systems. Codes mobilize signaling materials from languages, kinesic and proxemic systems, and probably from other systems we are still unaware of. More precisely, codes transduce communicative intentions into sets of multimedia signals, and then other codes translate these sets of signals into inferential interpretations of intent. But codes are not these materials themselves, nor the intentions themselves, nor the interpretations themselves: codes are the general functions that, only in the presence of a human mind, assign relevance to linguistic and other types of signals.

     To summarize, on the one end of interaction we have biological states of the mind/brain, and, on the other, our interlocutor's biological states of the mind/brain. In each of the interactants, these mental states are first cooked into notional or semantic representations. Here "semantics" must be understood as a very broad term: as a dimension that deals with content in a very general sense. These semantic representations include referential and propositional contents, beliefs, values, dispositions, affect, shared knowledge, etc. And here is where communicative codes emerge, as organizing devices to make sense of this cognitive ooze.(2)

     Those of you who are masochistic enough to have read some Umberto Eco may think that I'm simply triturating his complex model of "codes" -- specifically, his distinction between "s-codes" (or "systemic codes") and "C-codes" (The Code par excellance) (Eco 1976). S-codes are formal, syntactic or semantic, structural, non-interpretive. C-codes, for their part, are interpretive functions. In my view, Eco's "s-codes" approach "syntax" and "linguistic semantics", that is "language", whereas "C-codes" approach the communicative codes I'm talking about.

     Well, more or less. I say "more or less" because I've never understood Eco very well. I've always better understood Roman Jakobson's pioneering recycling of the notion of "code" from communication theory, that is, his rich and at times contradictory attempts to fit the mechanics of coding in communication theory into the mechanics of human communication through language (Jakobson 1959; Jakobson 1961; Jakobson, et al. 1952). In this sense, Jakobson was adamant in emphasizing that linguistic analysis should work "from the top down", that is, from semantics to sounds. (3) But, as we know, Jakobson was still very mechanistic in his communication model.

     So I believe I now understand why John Gumperz started to warn us a long time ago that what we should be looking at is "conversational inference" (Gumperz 1982). Again, neither can I assure that I fully understand what "conversational inference" is. But, if there exists a general mechanism in conversational inference, perhaps this mechanism consists of a bunch of communicative, interpretive/inferential codes. It is no coincidence that a central pillar in Gumperz's work is the study of code-switching, or the pervasive, ever-present, significant alternation of contrasting communicative codes in discourse. Now, this switching of codes may entail the display of materials from different speech varieties -- but then it may not. So even though most researchers of code-switching don't know this (yet), code-switching has very little to do with "languages": it has to do with the contrasting ways in which we contextually organize our cognitive ooze into diverse sets of communicative signals. (4)

     To return to Jakobson's initial question, what can we say about the sharing of codes in the interlocutors' minds? How do the coding processes work? First, as I have said, intentions about the communication of the sender's notional representations are coded into sets of signals. Then, these sets of signals are not "decoded", but coded again by the receiver into his or her own notional representations of the sender's intent. So I would like to venture that in human communication "decoding" never takes place. Speaker and hearer use not only different codes, but codes of a different nature. That, and not merely linguistic or cultural commonalities and differences, would explain both understanding and misunderstandings.

     Finally, then, which types of codes are there? How many codes function in a given human encounter? I don't know! But, if our semantic representations of reality are plural, then there should be multiple codes to handle this plurality. We should be able to find at least situational codes that deal with the signaling of our understanding of a given social situation and of our intentions about social identification.(5) Secondly, we may find activity codes, which are procedures that organize the signaling of intentions about whatever discursive activity is being carried out. And thirdly, we may find speech-act codes, which operate directly over specific illocutionary intentions (requests, commands, etc.). These types of codes work simultaneously and dynamically with whatever signaling resources humans have. And this simultaneous and dynamic interplay between diverse types of codes would explain, precisely, the apparent displacements in language indexicality.

     Needless to say, this tripartite taxonomy of codes is useless, and it would be totally preposterous to think that these three types of codes are the types of codes that do exist. Simply, number three is nice, and we have found that we can at least talk about three things: situations, activities, and speech acts.

     For the same reason, it would also be grossly excessive to conclude that any code "exists". Codes are, or should be, our constructions. The interesting thing is that, unlike languages, codes are not visible entities, but abstract formulations about the relationships between intentions, interpretations, and observable communicative signals. In this sense, a code is nothing but the coherent explanation of a given aspect of conversational data. So, how can we unveil the codes that operate on a given encounter? Simply, by constructing them and by assigning them a coherent form. The form of the code is the content of the data.


Eco, Umberto.  1976. A theory of Semiotics. Bloomington: Indiana University Press.

Gumperz, John J. 1982. Discourse Strategies. Cambridge: Cambridge University Press.

Jakobson, Roman. 1953. "Results of a Joint Conference of Anthropologists and Linguists". Supplement to the International Journal of American Linguistics 19(2), 11-21.

Jakobson, Roman. 1959. "Overlapping of code and message in language". American Anthropologist LXI(5), 139-145.

Jakobson, Roman. 1961. "Linguistics and communication theory". In Jakobson, Roman (ed.), On the structure of language and its mathematical aspects. Providence (R.I.): American Mathematical Society, 245-252.

Jakobson, Roman. 1984. "El lenguaje común de antropólogos y lingüistas". In Ensayos de lingüística general. Barcelona: Ariel, 13-33.

Jakobson, Roman, C.Gunnar M. Fant and Morris Halle.1952. Preliminaries to speech analysis: The distinctive features and their correlates. Cambridge (Mass.): The M.I.T. Press.


1. I'm translating from the Spanish version, in Roman Jakobson (1984), Ensayos de lingüística general.

2. Perhaps the process which translates brain states into semantic representations is also a sort of 'organic' coding process, but at the present state of research we don't know how this coding process could possibly work. So, we are left with semantic representations as the actual ends of the communicative coding process.

3. This cautionary advice was largely forgotten by generative syntax, and that's why a generative approach to language will probably never be able to understand what human interaction is about.

4. To insist on John Gumperz's work, these sets of signals include what he has called "contextualization cues": linguistic elements, prosodics, paralinguistics, kinesics, proxemics, and, again, perhaps other systems that our microscope hasn't yet found.

5. These situational codes may sound similar to the notion of "frame", but I view a "frame" as a cognitive structure, whereas the code is the function that relates elements from this structure to communicative signals.


Última alteração: 18-09-2003