- 'Speaking Our Minds' Book Club
Combinatoriality and codes
I read this book as part of an interdisciplinary reading group at Cardiff. As we found, there’s a lot to agree with in the book, but the commentary below focuses on two points that we found confusing.
Combinatorial communication
Chapter 2 claims to be about the impressive expressive power of language: we can construct an infinite number of sentences, expressing new ideas and capturing a huge range of meanings with a finite set of building blocks. At least, this is what Pinker and Fodor and others find extraordinary about language. The apparent target explanation of this chapter is whether code communication or ostensive communication is the most likely route to it. But instead of trying to explain the productivity and expressivity of language by focusing on compositionality and systematicity, the chapter focuses on ‘combinatorial’ communication of a particular kind. However, this neither fits with usual definitions of either combinatoriality (combining meaningless units) or compositionality (combining meaningful units where the meaning of the whole is composed of sub-meanings of parts), and I think ends up answering a very different sort of question.
So, combinatorial communication as defined in the chapter is where two (or more?) meaningful signals are combined to form a new signal whose meaning is not the sum of the meanings of its subparts (p. 27). Instead, a new signal is formed with a totally different meaning. So, adding monkey ‘pyows’ (leopard!) to ‘hacks’ (eagle!) results in a ‘pyow-hack’, which means that the group will soon move to a new location. This is not therefore a case of compositionality or combinatoriality. Scott-Philips admits that this is not really a case of ‘combination’ either (p. 50), since this is effectively just adding two holistic signals to form another distinct holistic signal, rather than a ‘combinatorial’ signal.
As far as I can tell, the most plausible way to interpret the claim in this chapter is that it is difficult to combine signals to (non-compositionally) form new signals, if there is not something obvious in the environment which correlates with the meaning of the new signal (so for which the new signal cannot be either a cue or a coercive behavior). This means that building up new, non-compositional, vocabulary by adding together existing signals is unlikely under code communication.
This seems plausible… but it’s hard to identify just what the question this chapter is actually addressing. The massive flexibility and expressivity of language has very little to do with whether we can add existing signals together and get a signal with a totally different meaning. Instead, it is usually taken to be related to our ability to add signals together to form new signals whose meanings are composed of the meanings of its sub-parts, and how we can add those together further in systematic ways to form e.g. sentences where again, the meaning of the whole is at least strongly related to the meanings of its sub-parts. What Scott-Philips instead seems to be addressing in this chapter is a form of vocabulary building – how to get new holistic signals from existing holistic signals, whose meanings are all unrelated.
The section on ostensive communication also seems aimed at vocabulary building – if you’re good at ostension, then you can come up with signals for whatever meaning you want. Here though, instances of "combining" (e.g. p. 43) really are instances of compositionality – new signals are generated whose meaning is composed of the meaning of its subparts. Here then, a) we are not always comparing like with like, and b) the role of both codes and ostension are related specifically to vocabulary building, not directly to the compositional nature of language.
So, while vocabulary building probably is easier under ostensive communication rather than the code model, and so ostensive systems might be more expressive in this sense, the compositionality of language and so its impressive productivity is not addressed. Further, to the extent that proto-language demands compositionality (does it?), this then does not show that code communication is inadequate to build proto-language.
The relationship between ostensive-inferential communication and communication via codes
This is obviously supposed to be the theme throughout the book: that you cannot get to language via an elaboration of the code-model, even by adding in ostension. Instead, the claim is that in some sense, ostensive communication is primary, and made more expressive via conventional codes (e.g. p. 16, and elsewhere). However, I found it hard to track what exactly this claim amounts to throughout the book (is it a claim about the actual development of linguistic systems? a conceptual claim? just a plea for a shift in research attitudes?).
The best I re-construct the claim is as this: code communication with ostension tacked on is not as flexible as language actually is; to get real flexibility, you have to start with ostension first, which is helped along by conventional codes. Given the claims made throughout the book that language is not code-like anyway (using ostension you can get words to mean whatever you want), this seems like a straightforward statement. But I do wonder how this plays out in the actual development of early linguistic systems. There it seems less straightforward.
First, the ‘code-like’ features of language are incredibly useful. Even the metaphorical or more flexible uses we put language to are often based on nets of semantic associations. Further, while we can in theory use words in radically flexible ways, a huge amount of communication does rely on meanings being fairly stable (conventional codes). This is because codes not only make communication more powerful, they also make it (cognitively) much easier.
This seems particularly relevant in the development of early linguistic systems, where, given that the first forays into ostensive communication were likely to be hit and miss, you’d need all the help you could get. Communicating with someone in the absence of a shared language is hard work, even in fairly simple contexts, and even if you both have A+ mind-reading abilities. What you need are ways of minimizing ambiguity to a level where mind-reading has a reasonable shot.
Accordingly, one way of minimizing ambiguity in communication that is often discussed in language evolution is iconicity (e.g. here in Chapter 5). Iconic signals ‘look like’ the things they represent, which should make it easier to grasp their meaning. Discussions of the important role that iconicity may have played in early communication systems is predicated on precisely the idea that linguistic communication is hard, even with ostension, so signs probably started off iconic and then later became arbitrary.
Another way of minimizing ambiguity is of course with codes. It then seems reasonable that at least some conventional codes were likely to have been derived or adapted from existing natural codes, with others added on via different means (e.g. perhaps conventionalized iconic signs). In this case, there would have been some ‘continuity’ (p. 48) between earlier code communication and later ostensive communication. Indeed, one of the questions that kept coming up in the reading group was what happened to earlier natural codes – surely hominids would not just drop them entirely, but, as great apes do, use them in ever more flexible ways. Something like this reliance on codes is also found in e.g. Section 5.4 – where proto-language includes a set of “more-or-less stable communicative conventions” (p. 117).
In this case, early language users would presumably have relied on, among others, both complex coding/decoding mechanisms (association) and mind-reading abilities (metapsychology) to get early proto-linguistic systems off the ground – not just one of them in isolation. On a conceptual level, and rather obviously, all you need for ostensive communication is ostension, but practically, associative mechanisms would also have been crucial in getting linguistic ostensive systems going.
However, it is hard to tell if this picture where both mechanisms have a role to play, and neither route (pure code or pure ostension) in isolation looks particularly plausible, amounts to a challenge to Scott-Philips’ view. Perhaps the difficulty in identifying claims here (and elsewhere in the literature) is based on ambiguities in the explanatory roles that particular mechanisms are supposed to play, and in exactly which stage of the evolution of language. Clearly, if the primary marker of the difference between linguistic communication and non-linguistic communication is deemed to be the use of ostensive-inferential abilities, then these will play a key role in explaining the emergence of "language proper", but if linguistic communication is identified in some other way (e.g. displaced reference), then the focus may well be elsewhere.
Thom Scott-Phillips 25 June 2015 (04:54)
I am of course delighted to hear that Liz and her colleagues discussed SOM at length in their reading group, and I am pleased to have this opportunity to address the concerns she raises.
In her comments on combinatorial communication, Liz suggests that my attention is not focused on exactly the right topic. She points out linguists often stress the importance of two particular types of combination: combinatoriality (combining meaningless units), and compositionality (combining meaningful units into a whole, whose own meaning is a composition of the meanings of the component parts). I focus instead on what I call combinatorial communication: combining meaningful units into a whole whose meaning is not simply the sum of the meanings of the component parts. Liz believes that because I focus on this, rather than on combinatoriality or compositionality, it is at least somewhat unclear what my discussion tells us: “it’s hard to identify just what question this chapter is actually addressing”.
Let me therefore try to clarify. One of the main goals of chapter 2 was, in a sense, a negative one: to argue that a communication system that has anything like the sort of combinatorial richness that languages do is extremely unlikely to emerge if it is based upon natural codes. To make this argument I focused on what I call combinatorial communication. I did this for a variety of reasons, one of which is that combinatorial communication is, more-or-less by definition, the most simple way in which meaningful elements can be combined. My question was: how far can you get combining natural codes together? And my answer is: not very far at all. Certainly nowhere near anything that looks like a language. If I’m right about this, then the fact that combinatorial communication is not a typical category for linguistic analysis is essentially beside the point. You just can’t get there from here. You have to look elsewhere for a way to get a widely combinatorial system of any sort off the ground.
Let me put the point another way. Liz writes that “The massive flexibility and expressivity of language… is usually taken to be related to our ability to add signals together to form new signals whose meanings are composed of the meanings of its sub-parts, and how we can add those together further in systematic ways to form e.g. sentences”. This is true, but it misses the point. The combinatorial power of natural languages does increase expressivity, but this expressivity is parasitic on ostension and inference. Many animals can, I assume, combine things together. There is no reason to thank that it is computationally difficult. (This is, incidentally, why I explored experimentally the possibility that bacteria use combinatorial communication.) So combining things is nothing special in and of itself. It is only special when it sits atop ostension and inference. The usual view that the expressivity of language is due to combintorics is only partly true, and the part it misses is, I insist, critical.
This point leads directly to Liz’s question about the sense in which I claim that ostensive communication is prior to the linguistic code. She asks whether this is “a claim about the actual development of linguistic systems? a conceptual claim? just a plea for a shift in research attitudes?”. All three, is the answer. It is first a conceptual claim: the linguistic code just is a set of conventional codes built precisely in order to increase the expressivity of ostensive communication. But what I was also concerned to argue throughout SOM is that this conceptual point translates into, yes, a claim about the evolutionary development of languages: ostension must come first. And for both these reasons, and others, I do think that some shift in research priorities would be healthy for the language evolution community.
All these issues come together in Liz’s assertion that “another way of minimizing ambiguity is of course with codes”. Much of chapter 1 of SOM was dedicated to making the case that, when it comes to conventional codes – which is what languages are – this is not true. It is common in linguistics to believe otherwise, but a (the?) key lesson from pragmatics is that all linguistic utterances are ambiguous. As I put it in SOM, “as codes, languages are very defective indeed. In fact, they are wholly ineffectual” (p.17, italics in original). Put simply, the linguistic code is parasitic on the ability of ostension and inference to handle the inherent ambiguity of conventional codes. This is the reason why I claim that it is prior, in essentially all the ways that matter from an evolutionary point-of-view.
Liz Irvine 30 June 2015 (08:36)
Thom – thanks for your response, but I want to have another go at expressing some ideas that perhaps weren’t clear in my first post, on the relation between your conceptual and evolutionary claims about language, and how they link up to research attitudes.
So, your conceptual claim (following Grice, Sperber etc):
- Linguistic communication is not code-like, but relies on ostensive-inferential capacities of the communicating individuals.
Your evolutionary claim, which you seem to suggest follows directly from this (?), goes something like this:
- The evolution of linguistic communication directly tracks the evolution of ostensive-inferential capacities. No other species have these capacities. There is therefore a straightforward discontinuity between the communication systems of non-human species, and the linguistic communication of humans, because no other communication system is ‘language-like’. (E.g. “…nothing that looks even remotely like language can emerge prior to the evolution of ostensive-inferential communication.” p. 46).
So, you make an inference from the conceptual claim about a distinctive feature of language, to an evolutionary claim of discontinuity.
My question is what the claim of discontinuity actually amounts to, and whether it is warranted. The worry tracks Bar-On’s distinction between synchronic continuity/discontinuity (e.g. what is distinctive about language, compared to non-linguistic communication) and diachronic continuity/discontinuity (whether there is a clean evolutionary divide between non-linguistic and linguistic species) (Bar-On, ‘Must we go Gricean?’ 2013, pp. 342-343). As she points out, synchronic discontinuity is compatible with diachronic continuity, and this is what I want to press on here.
So, to repeat, the inference above moves from a claim about what is distinctive about linguistic communication, to the claim that linguistic communication is evolutionary discontinuous with other forms of communication (nothing else is language-like). But this requires a few other moves.
First, on a conceptual level, one needs to decide what counts as being language-like in an evolutionary setting. But this does not follow automatically from a conceptual definition of language, or identifying its core features. One can agree that ostensive-inferential abilities are essential and unique to linguistic communication, but still think that there are evolutionary pre-cursors that do not rely on these abilities that are ‘language-like’ in an interesting sense. Asserting that only full blown ostensive-inferential communication is like full blown ostensive-inferential communication is fine in a conceptual sense, but is a peculiar assertion to make in an evolutionary setting.
Second, on an empirical level, one then also needs to assess which (aspects) of non-animal human communication systems are language-like. This is an on-going project.
So, in order to sustain your inference, you need to assume/argue at least the following:
1) Ostensive-inferential communication requires 4th order meta-representational capacities, and without these, no communication system is ‘language-like’. But Moore has argued that Gricean communication may not require this degree of meta-psychology, in which case there are simpler pre-cursors to full-blown Gricean communication. Arguably, these would be ‘language-like’ in an interesting evolutionary sense.
2) There are no ‘interesting’ stages of proto-Gricean communication, and there are no other relevant features of language that might make a communication system ‘language-like’. This is not sufficiently defended. For example, Bar-On’s account of expressive communication as a relevant pre-cursor to language is dismissed in one paragraph in the book (bottom pp. 46-47); apart from anything else, this is arguably too quick. Further, the worry that there is no ‘detailed account’ of something mid-way between code communication and full blown ostensive-inferential communication is not sufficient to disregard the possibility that there can be one. The reason why detailed alternative accounts are not in abundance is presumably because of the ongoing popularity of e.g. Tomasello’s cooperative communication model and the focus on social cognition in general. Yet there is increasing evidence that ape social cognition is not a million miles away from human social cognition (controlling for socio-communicative environment), which in itself suggests that there is some interesting notion of continuity to work with here (e.g. Slocombe and Lyn’s work, Call 2011 review, etc). Also, the idea that (full blown) ostensive-inferential abilities are the only important or interesting features of linguistic communication with which to make judgments on evolutionary continuity is perhaps a bit odd. Things like displaced reference, flexible use of signs, shared attention, learning mechanisms, etc, seem relevant too.
3) The only relevant sense of ‘continuity’ is in terms of the means or mechanism of communication (e.g. via ostension and inference). But there is another sense of continuity that applies to the linguistic/communication system itself. So, on the assumption that early language use was hit and miss, early proto-linguistic systems would presumably have still used existing codes in flexible ways, presumably made a lot of use of pointing, and on top of this would also have made slow and laborious use of the ‘direct route’ to generating new signs with fairly fixed meaning. So, in terms of what was actually ‘in the lexicon’, there may have been at least some continuity in the early stages. This is the sense of continuity that came up repeatedly in the reading group at Cardiff, but is quite different to the one meant (I think). (This is also similar to Bart de Boer's comment).
Now, in one way of course, whether we say that the evolutionary trajectory of linguistic communication was continuous or discontinuous doesn’t really matter, as these terms are pretty vague anyway (and with a different emphasis, the same data can support both claims). But when it does matter is when it comes to research programs and attitudes. If, on the one hand, a strong claim of discontinuity is made, then this makes it easier to ignore features of communication systems that might actually turn out to more ‘language-like’ than previously thought, or to keep to a narrow conception of what language is and what its important features are. Of course the flip side is that if continuity is claimed, then we might dismiss genuinely unique features of linguistic communication.
So, while trying to shift research attention to the core importance of pragmatics and social cognition in understanding language evolution is a good thing (though, following Bart de Boer’s comment, this is already a major research area in non-Chomskyian studies of language evolution), it seems like this can be done just as well by emphasising ‘PRAGMATICS IS REALLY IMPORTANT!!’, rather than making additional and perhaps problematic claims about evolutionary discontinuity.
Thom Scott-Phillips 12 July 2015 (22:10)
Hi Liz. Thanks for the detailed response, and apologies for the delay in replying.
You write, correctly, that “One can agree that ostensive-inferential abilities are essential and unique to linguistic communication, but still think that there are evolutionary pre-cursors that do not rely on these abilities that are ‘language-like’ in an interesting sense… on an empirical level, one… needs to assess which (aspects) of non-animal human communication systems are language-like”.
I entirely endorse looking at non-human communication systems for pre-cursors of full-blown ostensive-inferential communication. In SOM I set out tests that would unambiguously identify a communication system as ostensive-inferential; but I did not, it is true, discuss what potential “mid-points” might look like, and how we could identify them. I nevertheless agree, without reservation, that this is an important conversation to have. Mathieu’s and Katja’s comments are good starting points. Having said that, I would like to stress that what we should be looking for here are behaviours that may be partially ostensive (i.e. are produced with an intention to make it manifest to an audience that one has an intention to communicate). Yet there is almost no comparative work specifically focused on ostension. Yes, there is work on intentionality (shared or otherwise), and this is relevant, but there is very little on ostension, properly understood. I would be delighted to see more work of this sort. (I do know that there is some work on ostension in non-human species, but this is a small subset of research focused on more basic aspects of social cognition, and it is dwarfed by research programs focused on the codes used in non-human primate and other communication systems. This is what I mean by “very little”.)
Let me turn now to the three points that you argue I need to assume and/or argue for, in order to sustain my arguments:
“(1) Ostensive-inferential communication requires 4th order meta-representational capacities… but Moore has argued that Gricean communication may not require this degree of meta-psychology, in which case there are simpler pre-cursors to full-blown Gricean communication…”.
As you know, I argued in SOM that ostensive-inferential communication is a richly metapsychological activity. You are right that Richard, among others, has made counter-arguments. But I could not rebut those arguments in SOM because the detailed version of them was not published at the time. In fact, the detailed version is still is not. It is true that parts of the argument exist in some of Richard’s papers, but the most explicit versions are still under review at the moment. So how could I rebut them?
“(2) There are no ‘interesting’ stages of proto-Gricean communication, and there are no other relevant features of language that might make a communication system ‘language-like’…”
As I said above, I would happily welcome proposals of this sort. That is: I would welcome substantive proposals of what partially ostensive communication would look like. There is, again, very little work of this sort. You are right Dorit Bar-On’s work is one exception, but as I said in SOM, I do not find it persuasive.
It may also be relevant to point out that I don’t think that an author has an obligation to address all arguments contrary to their own view. Richard made a similar complaint to yours, and I stand by what I said in response: I explicitly stated in SOM that the agenda was not to provide a comprehensive overview, but to make a positive case for my own arguments. (Perhaps this is a disciplinary difference in expectations: I cannot help but notice that, of the various book club commentators, it is yourself and Richard – both originally trained as philosophers – who have complained that SOM does not discuss other positions in enough detail.)
“(3)… in terms of what was actually ‘in the lexicon’, there may have been at least some continuity in the early stages.”
I agree. But this is not the sort of continuity I argued against in SOM.
One last point. You write that “while trying to shift research attention to the core importance of pragmatics and social cognition in understanding language evolution is a good thing (though, following Bart de Boer’s comment, this is already a major research area in non-Chomskyian studies of language evolution)”. I can’t agree. If this is a major research area, why does “pragmatics” barely feature in the indexes of edited collections on language evolution? There are widespread acknowledgements that pragmatics is important, but this is mostly lip-service. Consider, for instance, research on the codes used in non-human primate communication. This is much bigger area than comparative pragmatics and, critically, this research rarely if ever consider whether the codes that are being studied are natural codes or conventional codes. This would not be the case if pragmatics was taken as seriously as it should be.