Skip to main content
Loading Explore guides…
Evolution of Language · Acedra
Home
Explore
Psychology
Evolution of Language
Sign in to save
Chat
If YouTube asks you to sign in, open the video on YouTube.
Watch on YouTube
Reload player
Transcript
113 segments
Copy
Hide video
0:00
ADAM CROOM: Hey, everyone. Welcome to Introduction to Cognitive Science. I'm Adam, and the focus of our lecture today is on the evolution of language.
0:11
In the last lecture, we reviewed basic concepts from evolutionary biology and anthropology, including evolution by natural selection, evolution by sexual selection, genetic mutations, gene flow, and genetic drift. In the last lecture, we also looked at the evolution of the emotions and facial expressions. So our goal for today is to look at the evolution of language-- in particular, the evolution of syntax. And for today, we will adopt a comparative approach
0:44
by looking at this article, "Evolutionary Roads to Syntax" published in the journal Animal Behavior. So what is syntax? As we have seen in previous lectures on syntax, some theorists adopt a formal definition of syntax as a set of rules, principles, and processes that determine how well-formed sentences are formed from words and phrases in a language. And this is a common dictionary definition of syntax. However, other language researchers that adopt a more empirical and comparative approach
1:23
to the study of syntax often use a modified definition. An evolutionarily friendly definition of syntax is a set of principles by which meaning-bearing units can be combined into well-formed complexes. So on this view, words are meaning-bearing units and sentences are complexes of these units. And by adopting such a modified definition, we are able to study the use of meaning-bearing units and their complexes in the communication systems of other animals.
2:02
So how did syntax evolve? There are many views in the literature. For today, we will consider three views. The first is the computational hypothesis, which is the hypothesis that syntax evolved, along with and as a result of our increased computational capacity. So as our computational capacity increased and developed, so did our syntactic ability on this view. The next view is the surface structure hypothesis, where syntax evolved as a means to expand
2:42
upon limited lexical repertoire constraints. So on this view, in order to expand upon our limited lexicon for communication, we started combining those component parts to communicate additional things. By combining component parts, it allows us to expand beyond what those individual component parts in our repertoire we were able to use to communicate. And then the final hypothesis is the deep structure hypothesis, which is the hypothesis that syntax evolved along
3:23
with and as a reflection of our capacity for event perception and cognition. So as humans, we perceive the world and cognize the world in terms of an event structure, right? So when you watch a boxing match, you see a boxer or an agent perform an action on a patient or a recipient. All right, so our perception and cognition of events is structured in a way that is similar to the way that language is structured. So on this view-- the deep structure view--
4:01
as our perception and cognition of events developed, so did our linguistic abilities, right? So our syntactic ability sort of was parasitic on the perception and action components. All right, so let's go ahead and look at each view in turn now. So first, let's start off with the computational hypothesis. Some language researchers investigate the evolution of syntax by using tools from computer science and describing language in terms of formal or artificial grammars with increased complexity.
4:43
The idea is that computationally simple syntax-- for example, finite state grammars-- require fewer computational operations, and therefore fewer cognitive resources than complex syntax, for example, phrase structure grammar. In finite state grammars or simple syntax, the meaning of a sentence emerges from the relations of its adjacent words or in terms of serial word order. A limitation of finite state grammars or simple syntax is that it cannot explain the entire range of phenomena
5:23
in natural languages, mainly because there are also dependencies between non-adjacent words in natural languages that require a more complex phrase structure grammar. As we have had fun pointing out throughout the course, there are many instances of long range dependencies in natural language, such as in the formation of wh questions. So we see here this is an example in the top figure of a finite state grammar. And below, we have a phrase structure grammar. And here, finite state grammars are
6:04
based on adjacent relationships between words or based on serial order. We're going to do this step first and then do this step next. And we're just moving step by step with the finite state grammar. This can be used to account for a lot of linguistic phenomena. However, it won't be able to account for all of the linguistic phenomena that we find in natural languages. For example, we've seen, as we see down here, using the phrase structure, we see
6:38
that when we form questions, we perform a movement operation. And this is happening in a non-adjacent way. Right? This is not following near serial order. So to account for this type of linguistic phenomenon, we need a stronger apparatus than this finite state grammar. And the main idea from the computational hypothesis is that thanks to our increased computational processing, our increased computational powers, that we're able to not be limited to this, but rather we have access to this.
7:24
And to illustrate what a finite state grammar is as well as to offer some insight into the song production of a Bengalese finch, let's go ahead and look at this top figure for a moment. So the top figure shows a probabilistic finite state transition diagram of the song repertoire of a Bengalese finch. So this is how it's going to work. We're going to start over here on the left with the sequence aaa. And a, b, c, these are just going to be labels for different vocalizations or notes
8:03
for songs, like a song note. Okay? So first, we'll start off with aaa. Then the next vocalization is going to be a b. So it's going to be a different note or pitch. After b, the next sequence is going to be bcadb.
8:24
Okay? Then from here, with a 12% probability, the next production will either be hh or with 88% probability the next vocalization will be eekfff. Okay? So imagine we took this path first. So the next vocalization was hh. After that, the next vocalization is a b. And then finally, with a 37% probability, the next vocalization will either be an ilga, or with the 55% probability, the next vocalization will be an lga. Okay? And with an 8% probability, we might loop back and do this--
9:08
perform an hh and then do this again. So that's what these loops mean. So if we go here, we can go back and then do this again. So let's go ahead and take a couple complete paths through this probabilistic finite state transition diagram so we can get an appreciation of how this little model can provide us with a tool for understanding the variability and the variety of songs that a Bengalese finch can produce. All right, so we'll go through this cycle four or five times.
9:48
So we'll start off with aaa and then b, bcadb, hh, b, lga. And now we've completed one turn. Now I'm going to go back. So jaa will be my next production. Then aaa, b, bcadb, eekfff, adb, hh, b, hh, b, lga. We'll go through two more times. jaa, aaa, b, bcadb, eekfff, bhh, h b, ilga. And just one more time for fun. jaa aaa b bcadb eekff, f f adb hhb hhb lga. Okay? So that was the little song that I played for you as a Bengalese finch.
11:07
So we saw that although this is really cool, and we can use these probabilistic finite state transition diagrams to model bird songs of different kinds, the limitation here is that we're unable to provide movement operations-- like complex movement operations-- using this simple finite state transition diagram. We're going to need something like phrase structures to do these wh movements appropriately. Another issue or limitation with artificial grammar research
11:45
is that stimulus sequences are usually meaningless, simple tones. This is a deliberate choice so that the syntactic apparatus can be investigated in its pure state, uncontaminated by semantics. Although the logic of this methodology is clear, there are questions about the ecological validity of this approach. In other words, do the results from computational modeling or experiments on artificial grammars provide findings that are really applicable to our understanding
12:18
of natural language syntax in real ecological contexts? This is sort of just a general worry about when we do experiments in highly controlled settings, does this really reflect what's going on in the brain in real life environmental contexts? And it may or may not. And it's just something that we always need to think about. And also why investigating something like syntax from various approaches is helpful, so we can converge on the most reasonable explanation.
12:53
Okay, so when we're studying syntax, we might want to do a combination of both research on artificial stimuli in controlled settings, but then also do more ecologically valid approaches, like studying syntactic processing in more realistic situations.
13:18
So there is a debate about whether artificial grammar experiments reveal something relevant for evolutionary theories of syntax, or whether they're more informative regarding acoustic pattern recognition. We want to be able to distinguish between just acoustic pattern recognition and syntactic processing. Regardless of the debates, the main idea here behind the computational hypothesis is that only humans can deal with complex grammars
13:51
due to the limited computational powers of other animal brains. The evolution of syntax on this view is a direct consequence of the evolution of computational power required for syntactic processing.
14:12
Next, we have the surface structure hypothesis. This is another influential hypothesis in the literature about the evolution of syntax. And this hypothesis is based on the idea that syntax evolves as soon as lexicons reach their limits, either due to memory or production limitations. Nowak, Plotkin, and Jansen, for example, have argued that, quote, natural selection can only favor the emergence of syntax if the number of required signals exceeds a threshold value.
14:48
For example, imagine an animal that has only three basic signals or calls or words. And we'll just label these A, B, and C, but they could be anything, like bleep and boop, or whatever, right? Just imagine such an animal-- a basic animal that only has three signals, A, B, and C. The animal may be restricted to these three basic signals due to memory constraints or limitations. For example, the animal may not be able to remember more distinct calls than this.
15:25
Alternatively, the animal may be restricted to three basic signals because of production limitations. Luckily for us, we have fine grain articulation ability, so we can produce many different types of vocalizations. But another animal without the fine-grained motor control and articulatory capacity may only be able to produce a few different sounds, right? Like my dog, for example, can only produce a certain number of distinct sounds. So this is another limitation on calls,
16:03
that given your vocalization ability, you may only be able to produce three different sounds. Okay? So since this animal has reached the limits of its lexicon-- right the number of distinct meaning units in its inventory, A, B, and C-- in order for this animal to communicate more things, such as additional threats, social status information, and so on, this animal will have to start combining the elements it already has-- for example, A-B or C-A-- to communicate those new things. Right?
16:42
So here, the important insight is that if I only have three meaning units in my mind and I want to communicate more meanings than that, at some point, I'm going to start combining what I have to mean new things. So although intuitively appealing, the hypothesis is difficult to test, because it presumes species-specific thresholds, but as of yet, there's no clear theory on how that could be determined. So this is almost like an ad hoc or after the fact type of explanation.
17:20
And really what we want from a theory is something with predictive power too.
17:27
Nevertheless, the hypothesis predicts that in closely-related species, syntax is only present in species that have reached the threshold-- the species with the larger repertoires.
17:44
There is a long ethological tradition of studying the surface features of animal communication-- the way that species such as birds combine elements of their signal repertoire into sequences. For example, how birds combine notes and phrases into songs.
18:04
Studies on birdsong have revealed that signal combinations play a role in social interactions. Recall in the last lecture, we were thinking about why we have certain traits or why an animal has certain traits. And we asked, did this trait serve as an adaptive solution to some recurring problem in the environment and the ancestral environment, right? And if a trait, either physical or cognitive, solves an adaptive problem of survival, then this trait is selected by natural selection.
18:44
If this trait or cognitive module solves a recurring problem of reproduction, then this trait is selected for via sexual selection-- evolution by sexual selection. So recall that we were thinking about evolution by natural and sexual selection. So we could think about what purpose or function different traits and modules serve in an evolutionary context. And here what we see is that birdsong functions to attract mates and keep rivals away, and contains information about caller identity,
19:23
such as social status.
19:28
Further research has also investigated whether these calls have specific meanings and what these calls might mean. And we've found some very interesting results, which I will cover here in the next few slides.
19:51
In primates, combinatorial calls have been found in Campbell's monkeys, capuchin monkeys, Diana monkeys, king colobus monkeys, putty-nose monkeys, and squirrel monkeys. And we see examples of our friends here. Right? So this is a Campbell's monkey. This is an example of a Diana monkey. Over here we have king colobus monkeys. Here is an example of a putty-nose monkey. And this over here is a squirrel monkey.
20:29
We see there are combinatorial alarm calls that contain information about threats from predators. And we find this in Campbell's monkeys, Diana monkeys, and king colobus monkeys. We also find combinatorial contact calls, which contain social information about the caller. And we find this in Campbell's monkeys and Diana's monkeys, among others. So here, we see that there are combinatorial calls of two types, an alarm type call and a contact type call.
21:08
The alarm call communicates information about threats, and contact calls contain information about caller identity social information.
21:23
Research on male putty-nose monkeys has shown that they produce two alarm call types, pyows and hacks. A series of pyows are used for terrestrial or land-based threats, such as leopards. A series of hacks are used for aerial threats, such as eagles. Additionally, males can combine both calls into pyow-hack sequences, which carry different meaning, such as traveling versus foraging, unlinked to the meanings of the component calls-- terrestrial threats and aerial threats.
22:05
So what's cool here is we see that there's something a differential semantics here between pyows and hacks for putty-nose monkeys. Right? They're systematically producing pyows for land threats and they're systematically producing hacks for aerial threats and also systematically producing pyow-hack sequences for a different meaning, such as traveling. So I think that that's really interesting-- fascinating. The composed meaning appears to reside in the pyow-hack transition, regardless
22:49
of the number of the pyows and hacks. And here in this figure, we see a distinct alarm caused by Campbell's monkeys and Diana's monkeys for leopard-related stimuli and eagle-related stimuli. So on the left, we have leopard growls.
23:11
We see that here these are leopard growls and these are eagle shrieks. Here what's interesting regarding the monkeys is that we have Campbell's monkey leopard alarm calls here on the left. And on the right, we have Campbell's monkey eagle alarm call on the right. So look at the leopard call. It looks like this. And now look at the eagle call. It looks very different. That's the important insight here, is that the Campbell's monkey call for a leopard and its call for an eagle are very different.
23:48
And we can see the difference here. Just like my word for leopard and my word for eagle is different, in a somewhat similar way, the Campbell's monkeys are distinguishing between these two different types of threats. So something like a semantics here, which is a pretty cool. Similarly, we see the same thing for Diana monkeys. So on the left, we see a Diana monkey alarm call for leopards or land-based threats. And then on the right, we see the Diana monkey call for eagle or air-based threats.
24:23
And we see again that these are very different, right? So the important insight is that there's systematic differential signals for different types of threats for these monkeys or these Diana monkeys and Campbell's monkeys.
24:47
Interestingly enough, research has also found that white-handed gibbon songs were produced both as duets and to predators. Predator songs and non-predator songs were assembled in different ways from the same basic song units, with syntactic differences, particularly visible during the earliest parts of the song. This research suggested that recipients discriminated between the different song types, right? Songs for predators versus song duets with social partners.
25:25
Some progress has also been made with the study of syntax on pant hoot call utterances. Using machine learning and automated feature extraction, the study produced evidence for encoding of age, rank, identity, and context across the four phases of their calls. All right, so here we just have a figure where we have the four phases of calls-- the introduction, build-up, climax, and letdown. And we see that there's different types of information conveyed in different parts of the call.
26:11
As we can see in figure 1 below, the introduction and build-up phases include low amplitude signals that contain mostly color identity information, suggesting that they are directed at nearby individuals. In contrast, the climax phase is acoustically conspicuous and includes high amplitude signals that contain information about both caller identity and social status, low rank versus high rank, presumably targeting far away group members and neighboring communities.
26:49
This is a relevant finding, because in chimpanzees, decisions about whether to engage in intergroup conflict are largely based on attending to neighboring pant hoot vocalizations. Finally, pant hoots usually end with low amplitude letdown units, which inform nearby group members about the caller's forthcoming behavioral intentions or subsequent actions. For example, I'm feeding versus traveling. This is what I'm about to do. Callers can omit or remove one or more of the four phases,
27:28
allowing them to target specific audiences with specific information. And I think that this is really fascinating.
27:39
All right, and we see here we can look through here how to read this figure. We see that social status-- I think that this is really cool that they encode social status information in their vocalization. And that's what's indicated here, right? We see social status is encoded here in this climax phase of the vocalization. We have context encoded and an identity information, age. Very cool.
28:11
Figure 1 shows four phases of chimpanzee pant hoop vocalizations, the introduction, build-up, climax, and letdown. Adults usually produce pant hoop phases in this order, although one or several phases can be emitted. Different phases convey different sets of information, such as identity, social status, context, and age, as indicated by the top information flow panel, which is provided right here.
28:44
Cool. Other research on great apes focused on the combined use of different modalities. In this research, call-gesture combinations were rare, and only used in social interactions of very positive or very negative connotations. Gestures were often added when vocal utterances failed to achieve a desired social goal, as an expression of underlying persistence. Schamberg and colleagues found natural call combinations in the wild population of bonobos. Schamberg and colleagues found evidence
29:25
for call-gesture combinations used to disambiguate meanings in the captive group of bonobos. These studies demonstrate that call-call and call-gesture combinations exist, and therefore provide the groundwork for further research on the evolution of signal combinations in our closest living relatives.
29:55
Bonobos also produce different sequences of call types that vary depending on the perceived quality of a food source. Playback experiments confirm that listeners were able to attend to the different sequences and make predictions about what type of food the caller had found. For gorillas, grunts, grumbles, and hums can be produced in isolation or in combinations. When produced in isolation, grunts were produced by individuals resting in close proximity of each other,
30:32
whereas grumbles were produced during foraging. When produced in combination, grumbles appeared to lose their foraging meaning, suggesting that call combinations here may be used to mark social roles during communicative interactions.
30:54
Research has revealed a wide range of combinatorial structures. Callers can combine vocal structures into merged units to convey identity and event info, such as in Campbell's monkeys and Diana's monkeys.
31:11
So we see pyow-hack, right? We're just sort of merging two units. But callers also demonstrate suffixation by adding acoustically invariable units to three distinct alarm calls to indicate that danger is non-urgent. Suffixation is different than the first example here of merging, because remember here with the merged units, we took two units, right? Pyow and hack. We just merged them, pyow-hack, to produce a new meaning, like I'm traveling. Suffixation, though, is different,
31:54
because what we're doing is we're using the oo units. We're adding it to a call or vocalization. And this suffixation is somewhat similar to suffixation in natural languages, like we can add a suffix to augment the meaning of one of our words. I can say punch, which means present tense. And I can add ed to the end of that to indicate that it was in the past. Similarly, we can add a suffix here, oo, to a call. And that will augment the meaning of the call. So that's why we call it a suffixation.
32:41
Research by Ouattara, Lemasson, and Zuberbuhler found that male alarm calls are composed of an acoustically variable stem which can be followed by an acoustically invariable suffix. This group found that suffixation in this species functions to broaden the call's meaning by transforming a highly-specific eagle alarm to a general arboreal disturbance call, or by transforming a highly specific leopard alarm call to a general alert call. And there's an awesome open access article
33:24
that I'd like for you to check out with this link. So if you go to this article, there's some nice audio files where you can hear these different calls, which I think is awesome. But visually, we can also see the differences here.
33:40
So for example, we have this vocalization, a krak. Right? But then we can see that this is the vocalization here, krak. But then we have a krak with the suffix, oo. Right? So we have the additional element added here. Similarly, we have a hok, right? And here we have the added suffix oo here. Hok-oo krak-oo. And by analogy, this would be like punch, punched. Kick kicked. Right? We're just adding E-D or a suffix to these different vocalizations to augment or alter the meaning.
34:29
And what this is doing in the case of this study is that it's sort of generalizing the alarm call, so it's not as specific.
34:43
Very neat. Something like suffixation in nonhuman animals.
34:52
In other research, callers produced permutations of call sequences, ordered call deliveries, in alarm calls. So here, we're pointing out the different ways that we can combine calls. So we saw we can merge like pyow and hack for pyow-hack. We also saw that we can do something like suffixation, where we get a krak or a hok and add an oo, so it's krak-oo or hok-oo. Okay? Here, this is just pointing out that we can also sort of like reproduce larger combinations or call sequences.
35:27
Callers can produce alarm call sequences that refer to both predator type and location. And this is shown in black-fronted titi monkeys. Finally, Zuberbuhler points out that there is evidence for meaning being conveyed by utterances of varying lengths assembled into complex sequences in the seemingly hierarchical structure. So for this last part, Zuberbuhler, who's the author of this article, suggests that there are also cases in other animals in which meaning is conveyed by utterances
36:05
of varying lengths, which can be assembled into complex sequences which seem to demonstrate hierarchical structure. And the examples here are Campbell's monkeys, colobus monkeys, and Diana monkeys. So if we're looking for potential syntactic abilities in non-humans, this seems to be a promising area of research. Okay? And recall that one of the ideas from cognitive science was that humans are unique in their syntactic abilities and in their productive abilities. Right?
36:44
And what we're trying to discover here is how unique are we really, right? Is it the case that maybe Campbell's monkeys approximate what we have? Or how different is what they have compared to what we have? We're adopting a comparative approach here. What's the proximity or distance between us and our closest relatives? All right. So we looked at the computational hypothesis and the surface structure hypothesis. So the last hypothesis to consider for today is the deep structure hypothesis.
37:27
The studies reviewed so far have not focused on underlying cognitive processes. And so it remains possible that what appears as syntax is not linked to any complex cognitive processing. In humans, however, syntax is tightly linked with how events in the external world are perceived, structured, and mentally represented. Like I mentioned earlier, when I watch a boxing match or some event in the world, I perceive the event as not just a stream of phenomenal qualities.
38:08
It's not just a stream of consciousness. But it has an actor that's performing an action on a patient, right? My perception and cognition of events is structured in something like a way that a sentence is structured, right? Like, I see a cat that's sitting on a mat. And the cat is the agent. The action the cat is performing is sitting on. And the recipient or the patient is going to be the mat. And so I cognize the event in this structural way,
38:42
which seems to be captured in my linguistic construction as well. We see that the cat is the noun phrase, the mat is the noun phrase-- is what's being sat on. And then here we have the verb phrase-- sat on the mat. So we have the noun phrase the cat, we have the noun phrase the mat, and then the verb phrase, sat on the mat. And the idea here is that the way that we perceive and cognize the world, that sort of structural format, then sort of serves
39:22
as a basis for linguistic structure to duplicate-- in terms of structural organization in language, what is going on with the structural organization of our perception and cognition.
39:40
So humans have a natural propensity to decompose events into actors, actions, and patients to the effect that there is a correspondence between the structure of natural events and the structure of language. And here I color coded this for you. So we see N's V P's, or actors act on patients.
40:15
So sentences are structured in that they contain agents, a doer cause or experiencer; actions, a what; patients, targets, or beneficiaries, the to whom the experience or state of affairs happens to. Okay. So we have sentences, the cat sat on the mat. This is captured and organized linguistically in this phrase structure. But it's also captured and organized in a mental structure for me, when I witnessed an event of a cat sitting on a mat.
41:03
All right. And we can just-- I'm just pointing out some additional facts about the organization-- the labeling and organization of events. So our arguments usually have additional components, such as the manner or how something is done, how an action is performed or an instrument that is used. Additionally, arguments can contain information about location or origin or direction or time. And these are things that can be identified both linguistically
41:38
but also in terms of our perception and cognition.
41:43
Another useful description of events is in terms of predication, so the eagle attacks; modification, the large eagle; and coordination, eagles and leopards.
41:58
So just as our cognition allows for these descriptions of events, so does our language. Languages have means to express these event features so that they are evident to listeners, usually with specific syntactic functions. For example, to syntactically distinguish an agent from a patient, some languages use phonological case marking, while other languages use word order. So for example, we can use word order to indicate who the actor and patient is here-- the agent and patient.
42:37
So if I have Sarah kicked Amber, we know that Sarah is the agent and Amber is the patient based on that order. But we can switch the order. So now it's Amber kicks Sarah. And now we know that the agent is Amber and the patient is Sara, because we switched the order.
42:56
The hypothesis here is that during human evolution, these event-bound cognitive universals have become externalized and assimilated into the communication system. This hypothesis appears to find support from research on NSL, or Nicaraguan sign language, which has shown that deaf children will gradually and without instruction from adults develop syntactic structures in spontaneous sign language that enables them to encode the core properties of an event, such as manner and path of motion,
43:35
rather than referring to the entire event holistically. And recall that we reviewed this article on NSL in a previous lecture. So if you want to refer back to the findings, check out that lecture on sign language acquisition, where we discuss manner and path of motion events. Modern humans, in other words, appear to have a natural propensity to mark the key components of external events with syntactic features.
44:13
So how do animals perceive natural external events? There is some evidence from artificial language studies that marine mammals can be trained to discriminate agents from patients. Research on chimpanzee vocal behavior has also found acoustic differences in screams produced in different social roles when the caller was the actor or the patient in an aggressive act, which was discriminated by others. Human event perception, though, is more complex than just marking agents and patients.
44:51
Zuberbuhler suggests that complex event perception is likely to have evolved first before the evolution of natural language syntax, possibly as a consequence of increasingly complex social systems and associated brain enlargements. Zuberbuhler suggests that syntax is a mere byproduct of perceiving external events in decomposed ways, and of the ability to mark these components with communication signals. The human road to syntax may have built on this predisposition and completed
45:32
with the advent of unprecedented vocal control, allowing event perception to become linguistically encoded with grammatical functions. There is ongoing debate about whether animal data can contribute in meaningful ways to questions about the evolution of syntax in humans. One argument is that studies of animal communication are irrelevant, because the only relevant property of human syntax is this generative hierarchical nature for which there is no evidence in animal communication.
46:07
Some scholars believe that the generative ability of humans is achieved by a single mental operation called merge, which takes two syntactic elements, such as a noun phrase and a verb phrase, and assembles them to form a linguistic unit, a sentence. However, Zuberbuhler points out that the merged view of language is not universally accepted, even among linguists. For example, much of ordinary language use is based on accessing prefabricated phrases from a vast memory stock. Hey, how are you?
46:49
Great, how are you? In our everyday communication, we draw upon these sort of holistic forms of communication in our expressions with each other. And we saw that Lakoff pointed this out in a previous lecture on conceptual metaphor.
47:09
Although the retrieved utterances may be analyzed in terms of syntactic structure, we can look at, hey, how are you, and provide a tree diagram. Language users simply retrieve these as wholes that fit into appropriate communicative slots. Prefabricated expressions account for up to half of all phrases used in conversations, suggesting that evolutionary investigations of syntax should also focus on non-generative, non-hierarchical combinatorial systems, as frequently seen in animal communication.
47:52
All right, so we've witnessed the argument for merge in the linguistics literature. Now in this slide, let's go ahead and look at different levels of merge or varieties of merge.
48:11
Zuberbuhler suggests that a more productive approach than the previous approach by Chomsky, which suggests that humans are unique in their merge ability-- Zuberbuhler suggests that a more productive approach from an evolutionary perspective would be to distinguish between different levels of merge, with increasing generative capacity. So this is a suggestion all or nothing approach. Zero-merge systems operate only with individual items from the lexicon.
48:46
This has been the default view of animal communication for decades, that animal signals function as holistic units without combinatorial properties. So if I'm a certain animal, I just might have a single vocalization or just two or three distinct vocalizations. A, B, C, and that's it. One-merge systems have combinatorial properties that allow for the formation of two-unit expressions. So word-word merges. So A-B can now be A-D. Rather than just A, B, we can now do a one-merge A-B merge.
49:32
We just call this a word-word merge. A two-merge system allows for recursion insofar as merged expressions, word-word merges that form a basic phrase, can enter new merges with its own components. This requires more memory capacity. So two-merge systems can potentially generate an unlimited set of expressions, word-phrase merges, and are therefore truly generative. So here with two-merge, this allows for word-phrase merges at this level. And what we mean by a phrase is just a word-word merge.
50:14
So when we are able to take a word-word merge, we can now label that a phrase. And then if we're able to merge that phrase with another word, now we're at this level, a word-phrase merge, or at the two-merge system, which Zuberbuhler suggests is where we finally get to truly generative systems. Next we have a three-merge system. And these are characterized by the ability to merge already-merged expressions. So here we get phrase-phrase merges. And this requires even more memory capacity
50:53
than the previous systems. Zuberbuhler suggests that sentence formation in human language requires a three-merge system as subjects and predicates consist of merged expressions. So we've seen in our complex tree diagrams on our lecture in syntax that we can pack quite a bit into that NP, right? We saw, especially with prepositional phrases, we can pack a lot under that NP. Likewise, we can pack quite a bit under that VP. And this is just to point out that in order
51:35
to capture all of that, Zuberbuhler suggests that humans must have a three-merge system. Current evidence suggests that animal calling systems or animal calls go beyond the zero-merge system but stop at the one-merge system without any recursive applications. So Zuberbuhler argues that previous work that thought animals only have single calls and that's it, those views are false. But on the other hand, animals don't have natural language syntax either in the same way that humans do. Right?
52:16
Their syntactic abilities or their combinatorial abilities exist, but at the one-merge level, according to Zuberbuhler, and that they can take a pyow and a hack and form a pyow-hack, but they can't form sentences such as, the snow in winter in Gotham falls slowly, or where is the Batmobile driving towards, and constructions like that. Or Bane knows that Batman knows that Catwoman got away, or whatever. So it's because we have three merge systems that we can play around with larger structures,
53:01
as we have been doing throughout the course.
53:06
So are humans unique in having higher level hierarchical syntax to generate meaning? Most definitions of human language require compositionality. And recall that the principle of compositionality is that the meaning of the whole is determined by the meaning of the parts and the rules for their combination. In several theory papers, primary call systems have been analyzed in such ways, which has led to the conclusion that some systems, particularly Campbell's monkey calls--
53:42
Campbell monkey call suffixation and putty-nose monkey call permutations have weak compositional properties. Zuberbuhler suggests that the evidence for compositionality is even stronger for some birds, with empirical evidence for compositional structure in both Japanese tits and southern pied babblers. So the answer to our question, are humans unique in having higher level hierarchical syntax to generate meaning? Well, kind of, but not really. Right?
54:19
So humans are unique in that we are able to do this. We're capable of this level of complexity. But we're not unique in combining at all, right? In that the other animals, as we've seen throughout the lecture, are capable of weaker forms of combinations. So other animals besides humans are capable of some sort of combinatorial processing. It's just not what we have at a three-merge level. So this is going to be the sticking point-- what distinguishes us humans from other animals.
55:11
At least, what the data suggests so far is that we're unique in our hierarchical processing, especially with the long range dependencies which are not based on serial order relationships.
55:29
And what's fascinating for us as researchers-- because this raises as many interesting questions, right? Like it raises the question of, is there a gradation in the animal kingdom for zero-merge, one-merge, two-merge, and three-merge species? And if so, where can we place all the animals, right? For example, where do our southern pied babblers fit? Do they have one-merge or two-merge, right? Or are there any animals or non-human primates that have a two-merge system? Right?
56:05
Or even a three-merge system. I think that would be really awesome to investigate, right? The interesting thing for us as cognitive scientists is, what animals might approximate like a two-merge system? That would be really cool to find such animals. And recall that we want to investigate a wide variety of animals, not just birds or even primates, but even also like marine animals also. We see that whales and dolphins also are incredibly intelligent and have call systems of their own,
56:43
which are worthy of investigation.
56:49
All right, so we're just going to look at figure 2. We'll end the lecture by looking at figure 2. Figure 2 shows examples of simple and complex compositionality in animals and humans. Figure 2 section a, so this top line, shows the male Campbell's monkeys that produce krak alarms to leopards and hok alarms to eagles. But both calls can also be merged with an oo suffix to generate krak-oo and hok-oo alarms or calls. In playback experiments, suffixation has shown to be meaningful to listeners,
57:29
suggesting that it has an evolved communication function. This system may qualify as having limited compositionality, since the meanings of krak-oo and hok-oo are directly derived from the meanings of krak and hok plus the meaning of oo.
57:53
Figure 2 section b over here, this is compositionality in birds. Southern pied babblers, our little friend over here, produce alert calls in response to unexpected but low urgency threats and recruitment calls when recruiting conspecifics to new foraging sites. When encountering a terrestrial threat that requires recruiting group members in the form of mobbing, pied babblers combine the two calls into a larger structure. And playback experiments have indicated
58:31
that receivers process the call combination compositionally by linking the meaning of the independent parts.
58:42
And finally, figure 2 section c down here. This is compositionality in humans. This is our friend Charles Darwin down here. Humans are capable of producing both simple, non-hierarchical compositionality and complex hierarchical compositions and dependencies. So Bane knew that Batman knew that the Riddler got away, or the snow in winter in Gotham falls slowly, and where did the Joker run off to? We can create all these hierarchically complex structures with long range dependencies, which
59:31
our friends the southern babblers and our Campbell's monkeys are unable to produce. So we exhibit that three-merge system, which the Campbell's monkeys and southern pied babblers have not demonstrated abilities for. Okay.
59:57
So the take-home message is that humans do seem to be unique in their syntactic abilities as linguistic creatures. But what we want to point out in this lecture is that other animals do not necessarily have zero-merge systems, that they do also display some very fascinating combinatorial properties, which are worth further investigation.
1:00:32
So to summarize what we've covered in the lecture today, what we covered today is the evolution of syntax. How did syntax evolve? And we covered three important hypotheses today-- the computational hypothesis, the surface structure hypothesis, and the deep structure hypothesis. Remember that the computational hypothesis was the hypothesis that syntax evolved along with and as a result of our increased computational capacity. So as our computational capacity evolved,
1:01:08
so did our syntactic abilities. The surface structure hypothesis was the hypothesis that syntax evolved as a means to expand upon limited lexical repertoire constraints. So in order to communicate more things than what I am constrained to given my lexical inventory, we have to start combining what's in our inventory to communicate additional things. And then finally, the deep structure hypothesis was the hypothesis that syntax evolved along with and as a reflection of our capacity
1:01:46
for event perception and cognition. So as we become increasingly refined in our perceptual and cognitive abilities, we perceive and colonize events as having rich structures. Similarly, our linguistic syntactic capacity was derived or parasitic on that, and evolved along with our evolved capacity for event perception and cognition. There's many more views to consider. But these are the three views that we considered for today. I hope that you enjoyed this lecture on the evolution of syntax.
1:02:26
And I look forward to seeing you in the next lecture. Have a great day, and I'll see you tomorrow for our next lecture.
Here to help you learn
Sign in to keep a private conversation with your course tutor.
Ask me anything about the material…
Sign in to chat