I just saw your recent paper on universals (“General linguistics must be based on universals (or nonconventional aspects of language)”). One immediate reaction I had was that a lot (perhaps most) of the phenomena that keep typologists busy are not universal but rather things that show up in some but not all languages. Just two examples (you’ll find more of them among the WALS chapters): classifiers and tone systems. You could perhaps say that they are universal in a very weak sense, viz. that hey are manifested in languages that have nothing to do with each other. In most cases, we have very vague ideas about the causal mechanisms behind them. What I wonder is whether linguistics isn’t unique among human and social sciences in its focus on universals.
Thanks for starting this conversation, Östen! As for your last question: I don’t think that we are the only social science that studies universals. For example, economists seem to assume that their models are universal, and there is also a lot of universalist anthropology, going back to Adolf Bastian and Edward Tylor in the 19th century. The same probably goes for sociology, though I’m less sure about that. There’s also some universalist musicology (e.g. Savage et al. 2015), and some universalist religion studies. In anthropology, this field goes by the name “cross-cultural studies”.
Maybe linguists calls the field “universals” because we have been extraordinarily successful in tracing the phylogenies of languages – other fields envy us for that. So we can’t simply say “cross-linguistic studies”, because most comparative linguistics is of the phylogenetic sort.
Östen: I agree that researchers in other social sciences also deal with universals, but what I suggested is that they do not share our focus on them, or think, as some linguists have done, that finding universals is their main task. Thus, linguists have developed a conceptual apparatus around universals that I think is not found elsewhere and which actually extends the concept of universal. Try googling “statistical universal” or “implicational universal” and you’ll hardly find anything that is not about language. It appears that when anthropologists speak of universals they really mean something that is positively found in all cultures, without any statistical or implicational hedges. Consider the following Wikipedia quotation:
“In his book Human Universals (1991), Donald Brown defines human universals as comprising “those features of culture, society, language, behavior, and psyche for which there are no known exception”, providing a list of hundreds of items he suggests as universal.”
Martin: Yes, it’s true that “universals” is not a term that is commonly used in these disciplines, even when they make general claims. For example, Watts et al. (2015) only talk about Austronesian cultures, but there is an explicit claim about causality, and this implies universality.
Perhaps the reason we talk about “universals” more often than these other disciplines is that most of the time, linguists actually particular languages, and comparing them is so damn hard. And in the Chomskyan context, the hope is that somehow we get at the cognitive foundations of languages without studying many of them. Maybe it’s mainly because of Chomsky’s outsize influence that “universals” became such a big deal in linguistics.
Im not sure what exactly you mean when you say that some non-universal things “keep typologists busy” – in what non-universal ways would you say do classifiers and tones keep typologists busy?
My paper about “general linguistics” is more about g-linguistics than about universals, because the main point is that one cannot really do g-linguistics by studying a single language. It has often been said that this is in fact possible if one studies phenomena that cannot be acquired without negative evidence, and I agree: the poverty of the stimulus is a good argument for a biological component of our knowledge of language (but linguists almost never make that argument in practice). But what linguists do in practice is to throw some “framework” at some data, and what I’m trying to say is that this works only if the framework is claimed to be an innate grammar blueprint (as in traditional generative grammar, I mean pre-2000).
I would say that what general (= non-areal) typologists study is universals, so when they study classifiers, they necessarily study universal aspects of classifiers. Otherwise they engage in p-linguistics.
Östen: As a concrete example of “what keeps typologists busy”, I looked up David Gil’s chapter on numeral classifiers in WALS, to see what kinds of statements he makes about them. Several of them involve existential rather than universal quantification, e.g.
• “It is sometimes the case that in languages with numeral classifiers, their occurrence in the numeral-plus-noun construction is dependent on the choice of the numeral or of the noun.”
• “Whereas in some languages each noun is associated with exactly one classifier, in other languages there is a degree of flexibility with regard to the choice of classifier, permitting the differentiation of subtle shades of meaning.”
Maybe you would say that such existential statements belong to p-linguistics? But typologists make them all the time, it seems to me.
Martin: These typological non-universal statements are interesting because indeed, they are paradoxical. E.g.
“In some languages (OR: in Mandarin Chinese), the occurrence of a numeral classifier in a numeral-plus-noun construction is dependent on the noun.”
This looks like an existential statement, but I don’t think that it makes sense outside of a larger universalist context. If our goal were merely to describe Mandarin Chinese (e.g. for some highly specific applied task), then we would not need general concepts such as “a numeral classifier” or “a noun”. We would have Mandarin-specific terms like “Mandarin-NC” or “Mandarin Noun”.
But we do often make such existential-comparative statements, e.g. “Latin has four cases”, or “Swedish has a postposed definite article”, or “Japanese has five vowels”. These make use of comparative concepts, crucially. With language-particular categories, we don’t make existential statements (*Japanese has Rendaku, *German has strong adjectives, *Swedish has (the/a) supine).
But why would it make sense to make an existential statement using a comparative concept? I think that such a statement is interesting only as a prerequisite for a later universalist stage of the research. We know that universalist progress is slow, so we pay attention even to slow steps toward our universalist goals.
Östen: I still think you owe us a clear definition of a “universal” (the noun). I took another close look at your “general linguistics” paper. The closest you get to a definition is the last sentence on page 1 which begins “If one wants to make claims…”, but when I now tried to see what you are actually saying there, I realized that the sentence has an Escher-style structure. Briefly, it is not possible to see if “strong tendencies” is an alternative to “universals” or to “absolute universals”.
Martin: Thank you for reading my paper so carefully, paying close attention to the terms I use! I do that with others, so I certainly want to be held to those same standards. I must admit that I have not thought this through, but I should.
But at the same time, I don’t want to make a rigid distinction between absolute and statistical universals, or between statistical universals and tendencies – so if you inferred that there is no difference between all these concepts, that’s right.
Östen: So what you want to say was is that universals can be expressed by propositions of the form “For all languages x, p(x)” or “For most languages x, p(x)”?
Östen: So since only some languages have classifiers, you have to add a conditional clause to any statement about classifiers: “For all languages x, if x has classifiers…” to make it conform to the universal template.
Östen: The question is now if just any conditional clause is allowed here. If it is, it is easy to find universals that are non-trivially true of a single language. For instance, suppose it is true (which is quite possible) that Norwegian is the only human language in which the words for ‘three’, ‘tree’, and ‘wood’ (the material) have the same shape (tre). Then every property of Norwegian is a universal, since you can use the homonymy I just mentioned as a condition in the universal statement. This is obviously absurd, but it is not so easy to disallow it if you want to keep your statements about classifiers as universals. At the same time, I agree that the typological study of classifiers is at a more general level than the description of classifiers in a specific language. So I wonder if it’s a good idea to identify general linguistics and the quest for universals.
Martin: Maybe you are right that “general” is better than “universal”. Interestingly, just a few days ago, I came across an old (1970) paper by Witold Mańczak, who criticized Greenberg (1966) for framing his generalizations in terms of “markedness” and “universals” – Mańczak says that we should simply say “linguistic laws”. Maybe he is right – after shedding the “markedness” baggage (cf. Haspelmath 2006), maybe it is time to shed the “universal” baggage, too – and maybe you and Mańczak are setting me on a trajectory in that direction. (And it reminds me of Itkonen (1998) – maybe I should look at that paper again.)
But the key idea, when talking about universals of numeral classfiers and other similar phenomena, is that the antecedent of the implication should be sufficiently general to make it possible to test such a claim by looking at a substantial number of unrelated languages.
For classifiers, this may well be the case, because at least the label “classifier” has been used very widely, for languages of many different families and several continents. So we should be able to find enough languages for such a test.
For “tre/tre/tre languages”, this is definitely not the case (because there is presumably only one such language).
Now what is “a substantial number”, and what are “unrelated languages”? I do not know – but I think, for example, that it is probably not an accident that asymmetric differential A-flagging occurs in particular in referentially prominent nominals (see, e.g., this blogpost), even though this does not occur very often. But the spread seems to be such that it can hardly be due to accident.
I realize that “universal” is a bit odd when it refers to implicational universals (and I don’t know other fields that have such a concept), but there is something law-like here – something that other fields seem to call “dependencies”. Maybe the latter would be a better term.
In any event, I think that the confusions between g-linguistics and p-linguistics are so enormous in our field that it’s important to highlight the need for general statements to be based on general data.
Östen: I agree that it is better to use the term “general” than the term “universal”, and clearly there is a sense in which we can have a general theory of classifiers that would apply to anything that fits the definition of a classifier.
I would like to relate that statement to the question of the nature of clusters. When trying to map a domain such as tense-aspect, you can observe smaller or larger groups of grams that behave similarly to each other. For instance, the English Perfect and the Swedish Perfect are quite similar in most respects. We also find that these grams are similar to other grams such as the Swahili Perfect or the markers we have called “iamitives” (Dahl & Wälchli 2016). As long as we are speaking of cases like the first one, it is likely that we have to invoke the shared history of the languages in order to explain the similarities. Thus, we have not yet found a “general” gram cluster, just a “local” one. But when we find similarities between languages that neither are genealogically related nor have been in close contact with each other, and particularly if the number is great enough to exclude chance, the cluster has to be explained by factors other than historical ones. When I re-read what you and I have said above, I realize that you have basically expressed the same thought – if we think of languages with asymmetric differential A-flagging in referentially prominent nominals as a non-local cluster.
But the distinction “local vs. general” is a tricky one, especially if you assume that external factors (such as climate, size of communities etc.) can influence language structure. If languages spoken at altitudes above 3000 meters have special phonetic properties due to the lesser oxygen content in the air, there will be statements that apply to those languages only, but irrespective of the region where they are spoken. So the statements will be general, but not necessarily universal in the ordinary sense.
Martin: Yes, all that makes a lot of sense.
Östen: But I am still somewhat puzzled by your insistence on a categorical distinction between language description (p-linguistics) and language comparison (g-linguistics). I think I can verbalize my puzzlement by a question, as follows: You have recently asked me to review a chapter in a reference grammar (for your Language Science Press series), i.e. (I suppose) a contribution to p-linguistics. And one might assume that the reason you ask me to review the chapter on the verbal complex is that I have done work in this area. But my work is mainly aimed to be g-linguistic. So how can a g-linguist be helpful to a p-linguist?
Martin: That is indeed an important question, to which I know the answer intuitively, but you are right that I should spell it out. And in fact, I am planning to write a paper with the title “The structural uniqueness of languages and the value of comparison for language description”. I got the idea to write such a paper after Nikolaus Himmelmann’s critique of my papers on comparative concepts at the last ALT conference in Pavia. Nikolaus says that I’m wrong in saying that each language should be described in its own terms, and in saying that one doesn’t need comparison/typology for description. But that is a misunderstanding.
Given the widespread experience that typology helps description, what is the relationship betweenn g-linguistics and p-linguistics? I would say that it’s not that g-linguistics offers the universal building blocks that p-linguists can make use of, but rather
• knowledge of g-linguistics (of worldwide patterns) can inspire linguists to ask questions that they wouldn’t otherwise ask – this has been very impressively demonstrated especially in the Relational Grammar literature (in my perception – I’ve long been interested in passives, ditransitives etc.)
• knowledge of g-linguistics help linguists use transparent terms that rely on readers’ g-knowledge, such as “plural”, “dative”, “applicative”; these terms are never sufficient to describe a language, but they make it much easier to understand a description in practice.
Östen: Yes, this is of course true, but it is of course not just the terms themselves, but the concepts they stand for that are important – and then the question is what the relationship is between the g-linguistic concepts and the labels used in p-linguistics, and how they make it easier to understand descriptions.
Martin: Yes, exactly.
Östen: In addition to what I just said, I’ll give you an example to consider. One of my favourite p-linguistic works is Li & Thompson’s Mandarin Chinese grammar – every time I open it I find out some new and usually surprising fact about this language. This time what I found, however, was an example of how they use g-linguistics to explain a p-linguistic phenomenon:
“Every language has grammatical processes by which a verb, a verb phrase, a sentence, or a portion of a sentence including the verb can function as a noun phrase. These grammatical processes are called nominalizations. Different languages, of course, may employ different strategies for nominalization. In Mandarin, nominalization involves placing the particle de after a verb, a verb phrase, a sentence, or a portion of a sentence including the verb.”
It is of course correct that you need more than this to describe nominalization in Mandarin – and they spend close to 25 pages doing so, but the point here is that “nominalization” is not just some g-linguistic term that helps you understand the rest of the chapter, but the way they present it, it stands for a concept which (at least in their opinion) is universal in language and applicable to Mandarin. I am not sure you think this is a legitimate way of looking at things, though.
Martin: I think this is a good example of what I said earlier: “Nominalization” is a general concept, and the claim here is that all languages have some language-particular category that matches is. But I think it is clear that the language-particular category cannot be identical to the general (comparative) concept because the latter cannot be defined in language-particular terms.
But let me get back to your “tense-aspect gram clusters” (which you related to “homeostatic clusters” in your 2016 paper). They were evoked in recent paper that is critical of my distinction between comparison and description, Matthew Spike‘s paper “Fifty shades of grue“. I think what Matt overlooked (or somehow failed to take into account) is that one cannot find out about such clusters without having properly defined the relevant comparative concepts (e.g. questionnaire sentences). Maybe once we know for sure that a cluster exists, we can give a label to such a cluster, and then we can say that such a cluster is a natural kind.
Östen: Well, it depends a bit on what you mean by “find out about”. You can certainly find a set of markers with similar distributions in a parallel corpus without fully understanding the nature of these markers. And I am not sure that questionnaire sentences are really “properly defined comparative concepts”.
Martin: I would say they are, because they apply to all languages in the same way.
Östen: A questionnaire sentence is just a sentence that I have decided to include in my questionnaire, I do not need to understand everything about it. A questionnaire of the kind you find in my Tense and Aspect Systems (1985) not really fundamentally different from a text (say, the New Testament, or the Declaration of Human Rights) that is the source for a set of translations forming a parallel corpus. It’s just that the author of the questionnaire thinks that the sentences are in some way representative of the problem under study.
Martin: Yes, and I think that parallel corpora are excellent sources of data for language comparison. So I really hope that there will be more work of the sort that you did in your groundbreaking 1985 book and that Bernhard Wälchli did in his work based on Bible translations.
But to get back to the nature of the tense-aspect “clusters” that you have found: It seems to me that for the vast majority of phenomena in linguistics, we don’t seem to have any certainty that clusters exist. Maybe your tense-aspect research has advanced far more than the rest of the discipline in this way…
Östen: I suspect that this is because we don’t think of the phenomena in terms of clusters. And after all, the notion of a cluster is a very general one: just a set of entities that are close to each other in some way. Just to take one example, if you look at p. 76 in your indefinite pronouns book (1997), you find a list of 37 attested combinations of functions of indefinite pronouns. Each of these combinations would correspond to a cluster of expressions in different languages which share these functions.
Martin: Now I have a question for you: What is a “gram”? I have recently written about morphs, and also about “bound forms” (you commented on my proposed definition of “affix”). But what is a “gram”? When you introduce the term in Dahl (2000: 7) (the Eurotyp volume), you don’t give a definition, but you limit yourself to giving examples (Progressive in English, etc.). I was long happy with this way of doing linguistics, but over the last 12 years, I realized that we need good objective “yardsticks for measurement”. As generative linguistics is getting stronger and stronger, I have come to see its main defect in perpetuating traditions – and if we want to avoid that, we have to give clear definitions.
Östen: What is a gram? That is a good question. My idea has been that grams are the basic units of grammatical systems, that is, the entities that you need to postulate to describe a system, and which cannot be reduced to combinations of other units, in the sense that their properties are not wholly predictable from their parts. I think this idea comes close to the idea of a “construction” in construction grammar, but construction grammarians do not seem to have a very clear idea how to treat things that would traditionally be thought of as features or values of an inflectional category, which are not naturally called “constructions”. I admit that the notion of gram is not very precise. In Bybee & Dahl (1989) we said that “gram” is short for “grammatical morpheme”, but it now seems to me that this forces you to stretch the latter notion quite a bit. It seems that grams are more abstract entities than your morphs. You say in your paper that feet is a morph, but you also seem to acknowledge that it is a plural form, so I assume you need another level where my grams would land. But what would you call the English Plural (which you sometimes capitalize) ?
Martin: Yes, I’ve thought about these issues for a long time, and it was only last year that I came to a conclusion: The never-ending discussion among morphologists about “morphemes” and “realization by exponents” is based on the wrong presupposition that the general concepts of comparison and the p-categories of description must be the same. As you note, this seems to make good sense (because p- and g-linguistics is so intertwined in practice), which explains that it took me so long to understand it. I’ll get back to the larger point later, but if one starts out from here, then it becomes easier, because in general, there are only four types of entities that are necessary:
• construction relations (= relations between sister constructions)
• form-classes (= sets of forms that can fill the same slot in a construction)
The forms (that belong to the form-classes) can be minimal, and/or simple, and/or they can be complex in the sense that they consist of parts that have relations with forms in sister constructions. There are presumably also larger construction families (maybe with inheritance hierarchies), and I cannot specify limits on constructions at this point.
So what is an “inflectional category (with its features)” (or “feature” with its “values”, as Corbett (2012) would say)?
I think that “Plural Number” ot “Past Tense” in English or Swedish are form-classes, some of whose members are minimal (e.g. feet, sang), but others of which are complex in the sense that they have sister relations with the Singular Number and Present Tense form-classes.
Östen: I don’t know if this is a conscious choice you have made, but you seem to be striving for an “extensional” definition of concepts. Even if you speak of “classes”, the impression is that you are really thinking of sets – which is something that would make the late philosopher Quine happy, since he wanted to eliminate everything except individual objects and sets. But if we choose to speak of sets (in the mathematical sense), the consequence is that “Plural Number” is just a list of forms. And if we don’t like that alternative, we should be able to specify what that approach is missing and how to remedy it.
So one thought that comes to mind is that a description of a language which speaks consistently in terms of form classes will be a bit unwieldy, for instance if you want to describe something like the agreement between nouns and adjectives in Latin or German. I think that if you try this seriously you will find in the end that you need things like “Plural Number” as what you could call an abstract index that you put on entities in the grammar to account for the behaviour. This is a bit like “hidden nodes” in neural networks, I think, but I am not sure I understand these well enough. Otto Jespersen has a discussion in his “Philosophy of Grammar” where he actually speaks of grammar as an “intricate network” long before anyone thought of neural networks.
Martin: Well, I have not thought this through entirely… But maybe I agree with Quine, why not?
But to get back to “grams”: OK, they are more abstract than morphs, and I don’t know how to define grams, because the “grammatical morpheme” constructions are so varied. So maybe they are defined by their meanings: certain meanings such as “progressive” are by definition grammatical meanings, so whatever expresses them is a gram?
Östen: I’m not sure how you then define “grammatical meaning”. The problem with progressives (like many other gram types) that grammatical progressives are not well delineated from constructions of a more lexical kind. Also, ever since I started thinking of gram types (or category types), I have not been able to make up my mind if they should be (exclusively) defined by their meanings or not. Actually, I’m not sure that it always makes sense to speak of the meaning of grams. Some grams are mainly triggered by a certain grammatical context, and usually the choice between different grams depends on several different factors which are sčąemantic to a varying degree. (Think of the choice of case and number with Slavic numerals. Sometimes forms like the Russian čаs-á (‘hour, o’clock’) – which are only used after numerals whose last element is 2, 3, or 4 – are called ”paucal” but that is strange when they follow a number like ‘24’. It’s a bit far-fetched to say that the form means “a few”.)
Martin: Or maybe you could say that if a certain meaning is expressed EITHER by a grammatical marker (= non-root bound morph) OR by a nonsegmental grammatical process (e.g. tone change), then it is a gram. I still have the problem that while I understand that “a morph expresses a meaning”, I’m not sure how a process can express a meaning.
Östen: And I’m not sure what you mean by “process”. 😊 Obviously, the form drank can be described by saying that you take drink and change the vowel from i to a, but in a sense that is only metaphorical. In a similar way to when I was told as a kid that 9 is a 6 that someone has turned upside-down. Speaking in terms of choices is a less problematic approach. Thus, the choice of a certain tone pattern can indicate for instance a certain tense. (This can be troublesome. Many languages do not represent tones in writing which makes it impossible to study their grammatical systems in written corpora.) This is no different from marking chess pieces as black or white to indicate which player they belong to.
Martin: It’s the same problem as with zeroes: I can say that -s expresses plural meaning, but not that -Ø expresses singular meaning. There’s no doubt a “zero construction” (and Bybee has a great 1994 paper about such cases), but there’s no “zero form”.
Östen: Right, but what you have is again a choice between two alternatives: the form with and without the plural ending. You can think of it in terms of slots. Here is a non-linguistic parallel: The now somewhat scandalized Swedish Academy has eighteen members each of whom has a numbered chair. If a member dies or leaves the academy, their chair is vacant, or empty. There are no zero members, though, only empty chairs. In a similar way, you can speak of an empty slot, if the grammar gives you a choice between having a marker or not having it. But parsimony requires you to restrict speaking of slots when there is a principled choice (whatever that means).
Martin: Incidentally, what do you think about Boye & Harder’s (2012) definition of “grammatical status”? I heard Kasper Boye talk about these issues recently at Peter Harder’s retirement colloquium in Copenhagen. As you know, they say that grammatical elements are “discursively secondary and ancillary in relation to other linguistic expressions”. It seems to me that it’s actually quite similar to my definition of “grammatical marker” in terms of “bound non-root morph”, because “bound” is very similar to their “ancillary”. A root can always be the focus of attention, but non-roots are ancillary.
Östen: Yes, in fact what they say is also quite similar to ideas I had in my (2004) complexity book, where I devote an admittedly relatively brief section to the inaccessibility of grams to focusing and related phenomena. (Boye & Harder refer to this section somewhat ungenerously as a “remark”.) But I think that this may just be one aspect of grammatical status. Another aspect, which is also discussed in my book and which I think is important, is what I call “cross-linguistic dispensability”, that is, the fact that the meaning of a grammatical item in one language need not be expressed at all in another. Most likely, both these phenomena reflect the ancillary character of grammatical markers.
Martin: Many thanks for this interesting conversation, Östen!
Bybee, Joan L. 1994. The grammaticization of zero: Asymmetries in tense and aspect systems. In William Pagliuca (ed.), Perspectives on grammaticalization, vol. 109, 235–254. Amsterdam: Benjamins.
Bybee, Joan L. & Östen Dahl. 1989. The creation of tense and aspect systems in the languages of the world. Studies in Language 13(1). 51–103.
Dahl, Östen. 1985. Tense and aspect systems. Oxford: Blackwell.
Dahl, Östen (ed.). 2000. Tense and aspect in the languages of Europe. Berlin: Mouton de Gruyter. https://www.degruyter.com/view/books/9783110197099/9783110197099.2.351/9783110197099.2.351.xml
Dahl, Östen. 2004. The growth and maintenance of linguistic complexity. Amsterdam: Benjamins.
Dahl, Östen. 2016. Thoughts on language-specific and crosslinguistic entities. Linguistic Typology 20(2). 427–437. https://doi.org/10.1515/lingty-2016-0016.
Dahl, Östen & Bernhard Wälchli. 2016. Perfects and iamitives: Two gram types in one grammatical space. Letras de Hoje 51(3). 325–348.
Gil, David. 2005. Numeral classifiers. In Martin Haspelmath, Matthew S Dryer, David Gil & Bernard Comrie (eds.), The world atlas of language structures. Oxford: Oxford University Press. https://wals.info/chapter/55.
Haspelmath, Martin. 2006. Against markedness (and what to replace it with). Journal of Linguistics 42(1). 25–70. https://doi.org/10.1017/S0022226705003683.
Haspelmath, Martin. 2021. General linguistics must be based on universals (or general aspects of language). (to appear). https://www.academia.edu/42175669/General_linguistics_must_be_based_on_universals_or_nonconventional_aspects_of_language.
Itkonen, Esa. 1998. Concerning the status of implicationel universals. Sprachtypologie und Universalienforschung (STUF) 51. 157–163.
Li, Charles N. & Sandra A. Thompson. 1981. Mandarin Chinese: A functional reference grammar. Berkeley: University of California Press.
Mańczak, Witold. 1970. Sur la théorie die catégories “marquées” et “non-marquées” de Greenberg. Linguistics. De Gruyter Mouton 8(59). 29–36. https://doi.org/10.1515/ling.19188.8.131.52.
Savage, Patrick E., Steven Brown, Emi Sakai & Thomas E. Currie. 2015. Statistical universals reveal the structures and functions of human music. PNAS 112. 8987–8992. https://doi.org/10.1073/pnas.1414495112.
Watts, Joseph, Simon J. Greenhill, Quentin D. Atkinson, Thomas E. Currie, Joseph Bulbulia & Russell D. Gray. 2015. Broad supernatural punishment but not moralizing high gods precede the evolution of political complexity in Austronesia. Proceedings Royal Society B 282(1804). 20142556. https://doi.org/10.1098/rspb.2014.2556.