Here I review which aspects of human language are behaviourally unique, what empirical support exists for such conclusions and if analogues or homologues for aspects of language are present in non-human animals (herein referred to as animals). I begin with a brief on animal signals then examine signals that typify specific characters of human language. I then move into a discussion of the capacities of animals versus humans to conduct computations necessary for the component of language considered central to many research programmes in recent years, recursion, before finally offering some avenues for further work.
There is a historic divide between organismal biology and linguistics (Margoliash and Nusbaum 2009). The resulting difficulty in incorporating linguistic theories with a biological framework seems untenable given that language constitutes a form of behaviour. Dawkins’ (1982) famously described behaviour as forming the ‘extended phenotype’ so we would reasonably expect language and our understanding of it to fit into an evolutionary framework. This does not preclude novel human characters from explaining our species’ extensive capacity for language. However, this explanation should only be invoked with sufficient evidence given its lower parsimony compared to an explanation relying on quantitative evolutionary differences (Pinker and Jackendorff 2005).
Animal signals comprise morphology or behaviour that influences the behaviour of another animal (e.g. Endler 1993; Johnstone 1997). Signals serve many functions and may have many receivers, including unintended predator species. Dusenberry (1992) classifies signals as facilitating ‘true communication’ if adaptive for both signaller and receiver or ‘deception’ if maladaptive for the receiver. Some signals are inevitably transmitted, notably morphological appearance, whereas others represent a voluntary action from the signaller. This leads to further categorisation of signals that comprise incidental transmission.
Studies like those of Basolo (1990) on evolution of the ‘sword’ in swordtail fish allude to the importance of receiver psychology and biases in forming signals. X. helleri putatively evolved swords due to an ancestral female preference shown in X. maculatus. Marchetti’s (1993) observations of Phylloscopus males exhibiting territorial display behaviours like wing-bar or rump patches flashing shows the interaction of behaviour with morphology to refine a signal. This allows signallers to ward off intruders whilst not remaining conspicuous to predators. Anolis lizards also combine behaviour with morphology to signal, varying the former depending on the signal. Male A. auratus perform territorial ‘assertion’ and ‘challenge’ displays, involving vertical head movements and dewlap extension (Fleishman 1992). The former is public and long-range with an initial high velocity element to induce a ‘visual grasp response’. This focuses receivers on the signaller after which the display is lower speed. Close range challenge displays target intruder males therefore do not include an initial element as assertion displays do. Adding the phonotactic whine and chuck mating call of P. pustulosus to this list to provides an auditory example of a signal which is acutely tuned to exploit its receiver’s sensory apparatus – the females’ papillae (Ryan and Rand 1990).
These are examples but they exhibit some fundamental properties that many signals do. Signals are ritualized, facilitating recognition, and repetitive to provide redundancy. Significantly, the anolis behaviours show the capacity for signal producers to differentiate signals, in this case on the basis of intended receiver; modest categories emerge. We shall see a development of categorisation and its importance in the context of semantics. Categorisation of signals renders them digital improving fidelity; signal degradation or interference is less relevant when the signal must only be interpreted as one of a finite number of mutually understood inputs, e.g. as with phonemes. There are also qualities which contrast with language, like the uncoupling of physical constraints like those imposed on the frequency of the tungara frog call. Such signals do not compare naturally with language and so serve as a good reminder of language’s special place in communication but as we shall see, the unpacking of language has still afforded many parallels in nature.
What Is Language?
Like any behaviour, language must be qualified in order to be investigated and consequently a number of criteria have been set out which, in a sense, define language. Several are presented in Hockett (1960) of which many endure. Today the following are recognised as most important. Language is infinite due to its combinatorial nature wherein meaningless units – phonemes – combine into morphemes – words. These words and sounds are discrete (digital) and words have fixed meanings (i.e. language is discrete and semantic). Moreover, connections between words and their referents are arbitrary and words can be conveyed using many modalities, e.g. speaking or signing. The units of language can be learnt and crucially are culturally transmitted, though the ability to learn them is not (Pinker 1994; Pinker and Bloom 1990). Humans can use language in a productive way, using general patterns to communicate new ideas, relationships and more. These general patterns are grammar and make language syntactical. The final but crucial aspect of language is its recursive nature. Humans are capable of producing phrases which contain other phrases, sometimes leading to the existence of long-distance dependencies. This quality is most contentious in comparative studies and is of special interest due to its highlighting by Hauser, Chomsky and Fitch (2002). We shall explore recursion thoroughly after discussion of other aspects of language found in animal species.
Aspects of Language in Humans and Animals
Most of the above qualities, perhaps with the exception of recursion, have been shown to some extent in animals. Beginning with displacement, we can look to the waggle dance behaviour of bees (von Frisch 1967). Through this communicative system, these animals signal with reference to a potential nesting site outside of their immediate sensory experience. Additionally, symbol-trained chimpanzees have been shown to be capable of spontaneously recalling and reporting the location of hidden objects (Menzel 1999). These constitute evidence of animal signals exhibiting at least a form of displacement (physical).
Aspects of discrete semantics have also been demonstrated. Seminally, Seyfarth and colleagues (1980) showed that three different alarm calls were used for snakes, eagles and leopards by vervet monkeys, with these each having tendencies to induce appropriate defensive behaviours in conspecific receivers. The authors found that production of the calls were contingent on detection of the relevant predator and also observed that stimuli eliciting the calls were refined with age. In an elegant later study, Zuberbühler et al. (1999) confirmed the arbitrary and semantic nature of the alarm calls by playing Diana monkey alarm calls for different predators (priming) before playing predator calls themselves. When primed with an appropriate alarm call, i.e. one that predicts the right predator, monkeys did not subsequently call but when primed with the wrong alarm, they did showing the conceptual content of calls. The contrast of the alarm call to the predator call itself shows that acoustic criteria are not sufficient for call recognition and further attests to arbitrariness of signal and the sematic nature of this animal signal.
Evidence for cultural transmission and learning exists for multiple species as well. Noad et al. (2000) showed that song patterns of humpback whales vary according to population which in itself is evidence for cultural transmission (i.e. different ‘dialects’ are observed). Further, he showed potential for learning by documenting the invasion of a Pacific population with an small contingent of Indian ocean whales that induced the replacement of the local song with that of the migrants. Following on from the work by Seyfarth’s group, Schlenker et al. (2014) detected dialect differences in the alarm calls of Campbell’s monkeys living in the Tai forest and the nearby Tiwai island. They propose that these are divergent populations that have culturally diverged in language due to separation.
Finally, word order and the combinatorial aspect of language has been demonstrated in dolphins, which are capable of interpreting linear order to follow instructions such as ‘take the hoop to the ball’ and ‘take the ball to the hoop’. Sentences of five words were understood in different modalities, aural and visual (gestures). Different individuals were capable of acquiring different grammars too, with some learning subject-verb-object grammars and others the reverse (Herman et al. 1984; 1993). Similar detection of word-order is present in the studies of the bonobo Kanzi by Savage-Rumbaugh et al. (1993) in interpreting spoken English instructions.
Taken together, these studies provide overwhelming evidence for the analogy of all aspects of language that are easy to delineate with the exception of recursion. I would therefore find it difficult to argue that humans are unique in any of these manners. In the section following, we probe recursion using the instructive framework of grammar.
Grammar and Syntax
The study of language with the approach modern linguistics employs is thought to have had its inception in the 6th-5th century BC when the Sanskrit grammarian Panini wrote the Aashtadhyayi (Staal 1972). The text described Sanskrit language in terms of a set of algorithms that accept lexical input to output grammatical language. Panini’s tact was systematic and mathematical and analogises that of Chomsky (1957) with his production rules, which allow for the formation of grammatical sentences via production of terminal elements from non-terminal elements.
To give an example of a production rule, the non-terminal sentence may ‘produce’ a noun-phrase and verb-phrase (thus may be represented S à NP + VP, for instance). In turn, the verb phrase may ‘produce’ a verb and a noun phrase. Through use of such production rules one may categorise all language according to the constituency grammar of Chomsky (1957; 1965), where all strings may be analysed in terms of production rules with units forming constituents of a whole phrase. The left-hand side of these rules include non-terminals (as they can produce other linguistic components) and the right-hand side variously includes terminals, non-terminals and null elements (the latter allowing non-terminals to be eliminated).
If a production rule includes the same non-terminal on the left- and right-hand side, we encounter recursion; the function is capable of calling itself. For example, S à aSb where a and b are arbitrary phrases but S is a sentence typifies centre-embedding, a form of recursion (e.g. ‘[the girl [the boy stopped] was tall]’). If we allow a and b to resolve to null elements (a à ɛ; b à ɛ) then we are capable of producing right- or left-branching sentences of the type ‘Abby told Bertrand told Cybil…’ ad infinitum. These involve extra phrases being added to the right or left of the main clause and are much more common in natural speech than centre-embedded structures (more common to computer programming; Pinker 1994).
The FLN and Recursion
As mentioned, the importance of recursion was emphasised by Hauser et al. (2002). The influential paper distinguished between the faculty of language in the broad sense (FLB) and the faculty of language in the narrow sense (FLN). The FLB putatively contains the conceptual-intentional system, the sensory-motor system and the FLN. The former two systems refer to the mapping of ideas onto words and the capacity for language production and recognition respectively. By contrast, the FLN represents the computational system that is required for language. As such the FLN represents the capacity for productive language via grammar and results in language having the quality of discrete infinity. Therefore, recursion is the core and only necessary component of the FLN as it is sufficient to produce strings of infinite length.
The authors go on to propose three distinct hypotheses for the evolution of language, the first two stating that the whole FLB is either homologous to animal communication or derived and uniquely human. The final hypothesis which they advocate states that the peripheral components of the FLB are homologues of animal systems but the FLN is uniquely human implying recursion is too, as the only necessary aspect of the FLN. Hauser et al.’s position is difficult to accept because computational aspects that could form part of the FLN are observed in other animals and the core component, recursion, has since been demonstrated, at least in certain forms, in other animals.
Hauser et al. (2001) employed a previously used protocol (Saffran et al. 1996) involving limited periods of morpheme playback to test statistical learning capabilities of infants on cotton-top tamarins. They found that tamarins could separate phonemes (representing a form of parsing), detect sequence structure and do all this rapidly, at a rate not incomparable to 8-month old humans. Advocates of the unique-FLN hypothesis would cite evidence such as that of Marcus et al (1999) that corroborates the ‘poverty of the stimulus’ argument (Chomsky 1965). Such studies suggest that humans have an internal grammar to enable them to deduce the whole grammar of their language despite exposure to only a small and limited lexical corpus, the collection of sentences encountered by the acquisition of speech. Marcus et al. show abilities of 7-month old humans to generalise past the experienced lexical corpus of artificial grammars used for purposes of the experiment. They also exclude the possibility that these capabilities exploit simple local recognition rules (finite-state grammars) alone to distinguish grammatical and ungrammatical strings. For example, grammars based on the simple detection of two consecutive words beginning with the same consonant type (i.e. voiced versus unvoiced) were ruled out by getting infants to distinguish between AAB and ABB type grammars, where A and B refer to words that begin with different consonant types.
Adding to the evidence against primates having lower capacity for linguistic computation is the demonstration (Fitch and Hauser 2004) that cotton-top tamarins, and likely other primates, cannot process phrase-structure grammars. These are broadly grammars which impose constraints over the course of a phrase and often rely on long-distance dependencies, for example the Engish ‘if…then’ contingency. The protocol used here exposed cohorts of tamarins to either a finite-state, (AB)n, or phrase-structure, AnBn, grammar mimicking centre-embedding. The responses (orientation) were measured and found to be significantly higher for novel stimuli which violated the rules individuals had been exposed to but only in the case of the finite-state grammars. Tamarins showed no significant difference in orientation to phrase-structure grammars, even when the probe stimulus was limited to a minimal length (‘ABAB’ vs. ‘AABB’) to control for cognitive constraints like those on working memory.
Though I believe the above to be in concordance with the postulation of innate language acquisition (Chomsky’s ‘universal grammar’ theory), I do not believe that these studies weaken the argument that computational components of language such as statistical learning are present in other phyla, including primate species. In fact, presence of such abilities in primates gives implies the FLN is multi-faceted, with some parts being evolved and present in non-humans.
To add to this, the evidence for recursion being present in some animals is compelling. Gentner and colleagues (2006) published a novel study in European starlings in which they used an operant go/nogo conditioning procedure to train cohorts of individuals with phrase-structure and finite-state grammars using the rattle and warble motifs of the bird’s song. They were able to show significant grammatical recognition from both cohorts, demonstrating capacity in starlings for acquisition of the computationally-demanding PSG, training starlings to recognise strings of the type A2B2. They also compared novel PSG probes against totally ungrammatical (by either grammar) strings to control for the possibility that birds merely detected the FSG-consistent strings and treated all others as a separate group. The researchers went a step further to test recognition of A3B3 and A4B4 strings, simulating generative grammars. Starlings showed that they could generalise according to the rule AnBn when tested with longer strings. This is significant evidence for recursion in non-humans, at least in the sense of embedding.
Corroboration of these results comes from Abe and Watanabe (2011) who habituated Bengalese finches to stings of the type A-C-F, again simulating recursive syntax. They observed a significant decrease in calling in response to ungrammatical strings which did not fit the PSG dependency of having an ‘F’ to follow an ‘A’ with ‘C’ units embedded. They were able to extend the length of the string to AA’CF’F and observed similar distinction between strings by birds according to the grammar. They were even able to incorporate additional C elements without birds deeming the phrases ungrammatical unless they broke the rule of the A-F contingency.
Studies by Rey et al. (2012) and Wilson et al. (2013) further the case by respectively providing evidence that working memory constraints may be responsible for the lack of recursion in baboons and that auditory grammars with branching structures can in fact be acquired by monkeys (macaques and marmosets). The final piece of counterevidence to Hauser et al.’s recursion hypothesis comes from anthropology. Everett (2005) summarises a life’s study of the Pirahã language, endemic to an Amazonian people, thought to descend from the Wari. Everett’s analysis of their language results in many surprises including that the language seemingly lacks numerals, fixed colour terms, a present tense, has the simplest pronoun inventory documented (and even this probably imported) and crucially, lacks embedding. The discovery of a group of humans without recursion who instead use separate phrases to convey what would usually be embedded features seems to fell the recursion hypothesis. It provides neatly reciprocal evidence that shows not that recursion is present in animals, but that it can be absent in humans.
Conclusion and Further Research
I do not support the recursion-only hypothesis and believe that there is abundant evidence to contradict it from animal (in particular bird) studies and from Everett’s discovery of a human language without it. However, given the critiques mounted by many (e.g. Corbalis 2007) that recursion has not conclusively been demonstrated I would recommend further research using protocols that can explicitly demonstrate specific dependencies between morpheme pairs in a string, i.e. A1A2A3B3B2B1. In this way, embedding cannot be confounded with counting. If possible, I would also conduct behavioural experiments on the capacities for recursion in humans with impaired working memory. Thus one can test if cognitive constraints have an important role in facilitating this linguistic behaviour.
Abe and Watanabe (2011), Songbirds possess the spontaneous ability to discriminate syntactic rules Nat Neurosci. 14(8):1067-74
Chomsky (1957) Syntactic Structures. Mouton.
Chomsky (1965) Aspects of the Theory of Syntax. MIT Press
conceptual skills of dolphins. In H. R. Roitblat, L. M. Herman, & P. Nachtigall, Language and communication: Comparative perspectives (pp. 273-298) Hillside, NJ: Lawrence Erlbaum.
Corballis (2007), Recursion, Language, and Starlings, Cognitive Science 31 (2007) 697–704
Dawkins (1982), The Extended Phenotype. OUP.
Dusenbery, D. (1992), Sensory Ecology W.H. Freeman.
Endler J.A. (1993), Some General Comments on the Evolution and Design of Animal Communication Systems, Phil. Trans. R. Soc. Lond. B. 340 1292 215-225
Everett (2005), Cultural Constraints on Grammar and Cognition in Piraha, Current Anthropology Volume 46, Number 4
Fitch and Hauser (2004), Computational Constraints on Syntactic Processing in a Nonhuman Primate, Science 16 January 2004: Vol. 303 no. 5656 pp. 377-380
Fleishman L.J. (1992), The Influence of the Sensory System and the Environment on Motion Patterns in the Visual Displays of Anoline Lizards and Other Vertebrates Am. Nat. Vol. 139, pp. S36-S61
Gentner et al. (2006), Recursive syntactic pattern learning by songbirds, Nature 440, 1204-1207
Hauser, M. D., Newport, E. L. & Aslin, R. N.(2001) Segmentation of the speech stream in a non-human primate: statistical learning in cotton-top tamarins. Cognition 78, B53–64
Hauser, M., Chomsky, N. & Fitch, W. (2002) The Faculty of Language: What Is It, Who Has It, and How Did It Evolve? Science 298, 1569-1579
Herman, L. M., Pack A. A., & Morrel-Samuels, P. (1993). Representational and
Herman, L. M., Richards, D. G., & Wolz, J. P. (1984). Comprehension of sentences by bottlenosed dolphins. Cognition, 16, 129-219
Hockett C.F. (1960), The origin of speech, Sci. Am. 203 88-111.
Johnstone (1997), Evolution of Animal Signals (Chapter 7), in Krebs and Davies (1997) Behavioural Ecology 4th edition. Wiley.
Marcus, GF, Vijayan, S, Rao, SB & Vishton, PM. (1999) Rule learning by seven-month-old infants. Science 283 77-80
Margoliash, D and Nusbaum, HC (2009), Language: the perspective from organismal biology. Trends in cognitive sciences. 13 505-510
Marmoset Monkeys, J. Neurosci. 33(48):18825–18835
Menzel, C.R. (1999) Unprompted recall and reporting of hidden objects by a chimpanzee (Pan troglodytes) after extended delays. J Comp Psychol. 113, 426-34.
Noad, M. et al. (2000). Cultural revolution in whale songs. Nature, 408, 537.
Pinker, S and Bloom, P (1990) Natural language and natural selection. Behavioral and brain sciences 13. 707-784
Pinker, S. (1994), The Language Instinct. Harper.
Pinker, S. & Jackendoff, R. (2005), The faculty of language: what’s special about it? Cognition, 95(2), 201-236
Rey et al. (2012) Centre-embedded structures are a by-product of associative learning and working memory constraints, Cognition 123 18-184
Rose (2004), Neural Mechanisms and Evolution of Behaviour from Electric Fish, Nature Reviews Neuroscience 5, 943-951
Rumbaugh. D. M. (1993). Language comprehension in ape and child. Monographs of the Society for Research in Child Development, 58, (3-4, Serial No. 233).
Ryan and Rand (1990), The sensory basis of sexual selection for complex calls in the tungara frog, Physalaemus pustulosus, Evolution 44(2) 305-314.
Saffran, J. R., Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8-month-old infants. Science, 274, 1926-1928.
Savage-Rumbaugh, E. S., Murphy, J., Sevcik, R., Brakke, K. E., Williams, D. L., &
Schlenker et al. (2014), Monkey semantics: two ‘dialects’ of Campbell’s monkey alarm calls, Linguistics and Philosophy Volume 37, Issue 6, pp 439-501
Seyfarth, R. M., Cheney, D. L. & Marler, P. (1980) Monkey responses to three different alarm calls: Evidence for predator classification and semantic communication. Science 210, 801—803
Staal, F. (1972), A Reader on the Sanskrit Grammarians, MIT Press
Von Frisch (1967) The Dance Language and Orientation of Bees, Harvard University Press
Wilson et al. (2013), Auditory Artificial Grammar Learning in Macaque and
Zuberbühler et al. (1999) Conceptual semantics in a nonhuman primate, Journal of Comparative Psychology, Vol 113(1), 33-42.