CC BY
6© Rahimi, A|i, and IHamzeh Haghighi 2008 This open access article is distributed under a Creative
Overview article Commons Attribution 4.0 International (CC BY 4.0).
An emergenist perspective of language origin: the computational modeling of language evolution
Ali Rahimi1, Hamzeh Haghighi2
1 Kashan University, Iran
E-mail: rahimijah@yahoo.com
2 Kashan University, Iran
E-mail: Hamzeh_haghighi@yahoo.com
Received: 1 March 2008 ^ie^ng edit°r: Andny G KirilW p online: 1 June 2008
Accepted: 15 May 2008 '
Abstract
The developmental psychology, cognitive sciences, and neurosciences try to understand how language is acquired and processed by humans at present. The researchers in these areas are interested in language origin in order to inform their theories. In addition to these empirical studies, computer modeling has joined the endeavor in recent years. This paper will focus on two areas. That is, we are connected with ways in which the study of language acquisition can contribute to explaining language origin from an emergenist perspective, and how computer modeling as a new methodology can be used for such purposes. It discusses how the study of language acquisition can contribute to the inquiry, in particular when computer modeling is adopted as the research methodology. Two important features of emergence, heterogeneity and nonlinearity, are demonstrated in the model.
Keywords
emergenist; computational modeling; language evolution; heterogeneity; nonlinearity; language acquisition
For citation
Rahimi, Ali, and Hamzeh Haghighi. 2008. 'An emergenist perspective of language origin: the computational modeling of language evolution." Language. Text. Society 2 (1): e1-e14. https://ltsj.online/2008-02-1-rahimi-haghighi. (Journal title at the time of publication: SamaraAltLinguo E-Journal.)
1. INTRODUCTION
In recent decades, there has been a surge of interest in the origin of language across a wide range of disciplines. People are fascinated by language and love to talk and speculate about it. Whenever speakers of different languages and dialects get together, one of the favorite topics of conversation is the comparison between their different languages. Scientists, not different from other people, like to talk and speculate on the origins of language as well, and the field of language origin has seen a renewed interest over recent years. Different questions about the evolution of language can be investigated. When did language evolve? Which of our ancestors had language? Was it relatively late invention, perhaps as late as 50 000 years ago when Homo sapiens apparently first started to make artistic and symbolic artifacts? Or was it earlier and did Homo erectus or perhaps even the Australopithecines already have language? A related question is that how fast language has evolved? Which evolutionary pressures played a role and what factors determined that humans ended up with language, while other animals did not? Apart from historical events and circumstances, there are more general processes that determined the evolution of language. These can also be investigated. How much of language evolution is the result of purely biological evolution and how much of it is cultural? What other factors besides biological evolution of individual human can have played a role? What was the role of co-evolution between language and the brain and that of co-evolution between infants' learning abilities and parenting behavior? What is the role of self-organization and a process often encountered in complex dynamic systems. All these questions have indeed been investigated by different researchers.
2. Inquiry into Language Origin
The issue of origin of language is one of the most important and fascinating questions in our understanding of human nature. Many of the early inquiries are little more than just-so stories. There had been so many speculative conjectures by the time Darwin published "On the Origin of Spices" that the linguistic society of Paris issued a ban on publications about the origin of human language in 1866. Only in recent decades has the investigation of the origin of human language returned as a scientific and collaborative enterprise. Since the late 1990s, the interest in language origin has increased dramatically, and a wide range of disciplines are joining in the endeavor to construct a plausible picture for when, where and how language originated, and how it has evolved. Among these disparate disciplines, genetic and archaeological studies propose tentative answers to the 'when' of language origin. One hypothesis speculates that when automatically modern humans first developed, the genetic disposition for language processing was already present. The earliest human fossils discovered so far suggest that this occurred at least 160,000 years ago.
Comparative studies on animals and their means of communication inform us about 'where' human language may have stared. It was previously believed that language was the result of genetic mutations specific in humans and that there was no continuum between human language and other animals' communication system. However, many capacities which were considered human-specific for language have been found in other animals in varying degrees.
Historical linguistics shed light on 'how' language could have emerged by showing how languages changed in the past, as does research on the genesis of pidgins and Creoles, as well as on the development of sign languages in isolated communities. The phenomenon of 'grammaticalization', by which content lexical words change into function words has been found to be pervasive across these investigations, suggesting that the earliest forms of language had no function words or grammatical morphemes, and the complicated syntactic system evolved from simple lexical items throughout the history of language change.
In parallel to the study of language in the past, developmental psychology, cognitive sciences, and neurosciences try to understand how language is acquired and processed by humans at present. The interactions are mutual, with researchers in these areas being interested in language origin in order to inform their theories. In addition to these empirical studies, computer modeling has joined the endeavor in recent years.
3. Emergenism
There are two main approaches to understanding language origin. The first focuses on the biological bases: what are the physiological, cognitive, and neurological mechanisms for language learning and language use? While it is clear that there have to be some biological prerequisites, it remains to be seen how many of these are human specific and language specific. Pinker and Bloom (1990) argued that humans are born with a language faculty, also called a universal grammar (UG), as a result of biological adaptation specific to language and to humans. However, there has been a great deal of debate over the actual components of UG. While earlier proposals for UG were mostly concerned with syntax, dealing with a set of highly abstract principles and parameters, recently the focus has shifted to more concrete components of language, such as the conceptual system, speech perception and production mechanisms, and the ability to store and process a large number of symbols. A recent review article on language evolution by Hauser et al. (2002) has been very influential in these regards, but their hypothesis that recursion is the only language-specific aspect remains highly controversial.
The second research focus concerns the social and cultural aspects of language origin. This approach pays more attention to factors such as interactions between individuals, social structures, patterns of cultural transmission, and their effects on the process of evolution of language in the community. It is argued that language could have evolved from simple communication systems through generations of learning and cultural transmission, without new biological mutations specific to language. While the human spices may have evolved to be capable of learning and using language, it is more important to recognize that language itself has evolved to be learnable for humans.
The two approaches to language origin that we have outlined above find a parallel in language acquisition research and the long-standing opposition between nativism and empiricism, or between nature and nurture. In recent decades, emergenisim, according to MacWhinney (1999), has appeared to replace the traditional opposition with a new conceptual framework, explicitly designed to account in mechanism terms for interactions between biological and environmental processes. According to Ellis (1998), "language emergence in individual learners can be explained by simple learning mechanisms, operating in and across the
human systems for perception, motor-action, and cognition as they are exposed to language data as part of a communicatively rich human social environment by an organism eager to exploit the functionality of language."
Emergenism emphasizes the importance of integrating the two approaches: on one hand, we have to sort out the significant and necessary innate abilities in humans which enable language acquisition, and, on the other hand, we need to understand the environment's profound impact on the learners, the learning process as well as the end product of learning. Emergenism also provides the study of language origin with a framework for integrating the two approaches provided above. Language origin and language acquisition are both emergent, albeit at two different time scales: phylogeny over tens of thousands of years at the macro-level, and ontogeny over a few years at the micro level. These two levels of emergence inform each other.
It is highly unlikely that language could have sprung spontaneously from a group of early humans within one generation. Schumann and Lee (2003) argued that a full-fledged language should have agglomerated its complexity gradually over the course of many generations, which means that the learning of the younger generations must have played a crucial role in the process. The initial biological condition for language acquisition of humans today should be the same as, or at least very close to, that of humans at the time when language first developed.
If the initial condition for language acquisition is a universal grammar (UG) which is specific to language, the task for the study of language origin becomes to explain the origin of UG: why and how it was selected biologically. Recent research, however, has argued that language acquisition can be better explained as a lexically-based construction process. The initial condition of language acquisition, according to Tomassello (2003), may require only a set of general cognitive abilities, non-specific to language, such as symbolization, intention reading, pattern finding, imitation, and cross-modal association, etc. Instead of having a language instinct, humans are better described as having a communication instinct and an instinct for learning in general.
Moreover, if the initial condition for language acquisition is indeed far less than an autonomous syntax module, then the key to explaining language origin lies in examining the dynamic processes of emergence, instead of dwelling on the properties of individuals. This shift of focus of investigation is in line with a general paradigm shift in science since the mid-twentieth century.
Emergenism pervades the complex adaptive systems in nature and human societies: snowflakes, honeybee combs, termite bounds, schools of fish, flocks of birds and economies and ecosystems are all emergent phenomena. In these complex systems, the emergence of complex structures at the global level is explained as the result of the long-term iterative interactions among the individuals inside the systems. The individuals do not have innate knowledge or a blueprint of the global structures, but each performs simple actions with limited knowledge of the local environment without any central control. Many computer models have successfully demonstrated such process.
4. The Emergence of Word Order
The emergence of word order is now introduced as an example to illustrate how word order could have emerged. All languages organize words in a certain sequential order. Even in
languages which have rich case marking and more flexible word orders, such as Latin, there is still a dominant order. In syntactic theory, word order involves more than putting individual words in a certain order; word order entails rules of how categories of words should be put together. Therefore, the knowledge of word order presumes the existence of knowledge of syntactic categories. Nativists hypothesize that children have an innate linguistic knowledge about syntactic categories, and when their knowledge of word order is triggered by linguistic input, they are able to productively construct multi-word utterances from very early on.
However, this view has been challenged by many in-depth analyses of early multi-word utterances in children's speech data. It is argued that children acquire syntactic categories from generalization of early lexically-based constructions. Children's first multi word utterances are found to be holophrases imitated from adult's speech, such as 'I dunno', 'go-away', etc. whose internal structures are not recognized by children. Later at around 18 month, many children start to combine two words or holophrases, for example, 'ball table', 'baby milk'. Also, around the same age, many of the multi-word utterances appear as pivot schemas, such as 'more_and_it', where one event word is used with a wide variety of object labels. Tomasselo et al. (1997) demonstrated the productivity of such pivot schemas, as children can apply novel names to these schemas immediately after the names are taught. For example, when taught a novel object label 'Look! Wug!' the children were able to produce sentences like 'Wug gone' and 'more wug'. However, children at this age do not make generalizations across various pivot schemas, and they do not have the syntactic categories yet.
At a later stage, around 2 years old, children go beyond pivot schemas. They can understand 'make the bunny push the horse' which has to depend on the knowledge of word order. Also, they can produce utterances which are consistent with the canonical word order, as evidenced by utterances from over-generalization such as 'don't giggle me'. This type of over-generalization has been used as an argument for nativism. However, such errors are rarely seen in children's speech before about 3 years old, which suggests that the knowledge of word order does not come from the very beginning. Furthermore, Akhtar (1999) showed that children around 2-3 years old would correct an utterance which violates the English canonical order if the verb is familiar verb such as 'push', but they did not correct novel verbs such as in 'Big Bird the car gopping!' Interestingly, older children (4 years old) tend to correct word order to match the canonical order, which implies that by this age they have mastered the word order as an abstract syntactic structure.
The findings from language acquisition described above have led to a hypothesis for language origin, which suggests that language may first start from holistic utterances, from which words or phrases/schemas are extracted as recurrent patterns, and later used in combination to express new meanings. This hypothesis differs from the scenario proposed by Jackendoff (1999) and others which suggests that there is a one-word stage when single symbols, i.e. words, are used for communication, and later words are concatenated following some basic word orders.
5. The Use of Computer Modeling
In order to understand the use of computer modeling in the study of the evolution of language, we need to understand that there are two levels to language: the level of individual and
the level of population. These two levels interact and this is an important factor in what makes the dynamics of language in a population so complicated.
At the individual level, language is made up of individual speaker's knowledge of the language, of their limitations in production, of the speech errors they make; of the way in which they acquire etc. This is the level that is related to what Chomsky has called performance. It is studied by what psycholinguistics who study such things as reaction times in retrieving words and limitations on short-term memory, by researchers of speech errors and speech pathologies, by researchers using neuro-imaging techniques and by researchers of language acquisition. Language at this level is intricately related to the functioning of an individual brain.
On the level of population, language is a conventionalized communication system, with a vocabulary and a set of grammatical rules. The knowledge in population is uniform to such an extent that users of language can communicate meaning and intentions with it. This is the level that is related to what Chomsky has called competence. It is often assumed that language at the population level is uniform over space and time. It is also often considered as an abstract system that exists in a sense separately from the individual speakers. Language at the population level is studied in historical linguistics and general linguistics and is also what is described and prescribed by language teachers.
Both perspectives are equally valid when studying language. It would be impossible to reconstruct the history of language if one had to take into account the behavior of every individual. It would be impossible to study organization of language in the brain without looking at the behavior of individuals. However it is obvious that these two levels do not and can not exist separately. The population level is an abstraction of the collective behavior of a group of individuals. Behavior on the individual level is influenced by what individuals perceive of the language used in the population of which they are part. The interaction between these two levels is a feedback loop. Changes in the behavior of an individual can change the collective behavior and this in turn can influence the behavior of individuals.
These feedback loops are by no means simple. The way language is learnt and the way innovations spread through a population is a complex processes. Such systems cannot be described in a mathematically simple way. In a technical mathematical sense they are non-linear systems. As has already been observed by Steels (1998) language is a complex (non-linear) dynamic system. The behavior of such systems is not easy to predict or even to describe. If one makes hypotheses about such systems, they will be extremely hard to test using pen and pencil alone.
This is where computer models come to rescue. One described in sufficient detail, complex dynamic systems can be implemented as computer models. Computer can then simulate the behavior of these models, and provide insights in how they work. When one compares the behavior of the computer model with the behavior of the real system, one can check whether the predictions of the theory correspond to what is found in reality or not. What computer models it would be extremely hard even to check what the exact predictions of the theory are. A common misunderstanding about computer models is that they only produce what has been put in beforehand and that they are therefore unable to produce any really surprising results. A complex dynamic system's behavior is so difficult to predict that the results of simulating it are often very surprising.
Another advantage of using computer models is that one can use them to do what-if experiments. When studying language evolution or other large and difficult to control problems, it is often impossible to do controlled experiments. It is possible to observe behavior of the system under study, but it is possible to change the initial conditions and see what happens or to restart the system to see what has happened in an earlier phase. Sometimes natural experiments happen such as when a pidgin or Creole language is formed but there are always many factors that one does not control. With a computer model, however, one has complete control over all parameters and even over the exact dynamics. One can also run or rerun the model as often as one wants. Computer model, therefore, make it possible to do as many hypothetical experiments as one wants.
In many fields of science, computer models are indispensable tools for investigating natural systems. One such field is meteorology, and more specifically climate modeling. The earth atmosphere and its ocean also form a complex dynamic system that would be impossible to understand without computer models. Computer modeling allows us to investigate long-term dynamics of this system and to perform hypothetical experiments on it by changing parameters and investigating how they influence the model's behavior.
Understanding how computer models are made and understanding how to interpret the results from computer models requires understanding of how an abstract system, such as a computer model, and reality map onto each other. Because computer power is limited, and because our understanding of language is limited as well, building a computer model requires us to make abstraction and simplification. This is not a problem. Abstraction and simplification are necessary for any scientific theory. Finding the right simplification is also the key to making successful models of other complex phenomena, such as the example of the climate as mentioned above. However, we should remain aware of the kind of simplification we make. It is very important not too simplify a model too much, and thus to remove all interesting dynamics. This sometimes happen in systems that are designed for mathematical analysis. Mathematical analysis can only be done on the simplest possible models, and the kinds of models we are interested in are generally not solvable analytically.
When interpreting the results from the computer models, we should be aware of how the results of computer model map onto the linguistic phenomena under study. For a model of speech sounds this mapping is usually quite straightforward. Such models work generally with direct representations of physical properties of the speech sound under study. For models of more abstract properties of language, this mapping can be quite intricate. Semantic can serve as an example. Meaning in computer models are often implemented as simple numbers that are a measure of how strong the association between a word and object in the world is. This is easy to implement, but a rather strong simplification of the complexities of semantics in human language. Such more abstract representations require an effort from the author to present the results of the model and the reader to interpret them. It is therefore essential to communicate the mapping between objects in the computer model and real linguistic entities and to explain how the results of a computer model shed light onto the real linguistic phenomena.
One should also be very careful not to use computer models to investigate aspects of language that they have not been designed for. For example, one can build a computer model for investigating language acquisition. Although this is a very obvious example, assumptions and
simplifications in a computer model can be extremely subtle. It is easy to forget the exact nature of these assumptions and the problem gets worse when a computer model that one researcher has designed is used by other researchers.
Deciding which abstractions and simplifications to use is one step in making a computer model. Another step is which computational techniques to use for the computer model. Sometimes the problem one is interested in and simplifications one has made already determine which techniques can be used. Like the abstractions and simplifications, all different techniques have their advantages and disadvantages.
Computer modeling is a widely used methodology in the natural sciences and engineering in order to simulate complex real world. It provides virtual experimental laboratories to run realistic, impossible and counter-factual experiments and tests internal validity of theories. In order to build a model based on chosen theories, the modeler needs to make all the assumptions in the model explicit and implementable. The models are usually highly idealized and simplified, so that a modeler can run controlled experiments on a number of parameters and different initial conditions, in order to examine their effects on the system behavior.
In some situations, models may seem circular: the modeler builds in what they expect to see, and therefore, the results are not unexpected. However, as Nettle (1999) pointed out, the interest in modeling does not lie in what the model can be made to do, but rather what assumptions and initial conditions have to be included to make the model produce the desired result. More importantly, there are times when the simulation leads to dead-ends or unexpected outcomes. Then, the modelers have to carefully examine and modify the existing assumptions and parameters. Modelers can identify new directions for empirical studies in order to address problems arising from the failure of the models. The beauty of modeling does not lie in producing results which confirm the hypotheses, but more in the process of building the model.
6. Computer Modeling Techniques
There are many different techniques that are suitable for modeling the evolution of language. Most of these techniques can be divided in three categories: optimization techniques, genetic algorithm and agent-based models. Optimization techniques define a quality measure on linguistic systems and try to optimize it. Genetic algorithms are techniques inspired by biological evolution that try to evolve a good linguistic system using a population of candidate solutions. Agent-based models model language users as simplified language programs and try to emulate how they use language. These categories provide a framework for presenting the different techniques, but it should be kept in mind that they are somewhat arbitrary.
7. Agent-Based Modeling of Language Origin
Agent-based modeling is a type of computer modeling which has been widely used and proved to be fruitful in offering new insights into the study of complex systems including man-made systems such as stock markets and traffic jams, and natural systems such as immune systems, ant colonies, etc. In an agent-based model, there is usually a group of individual components -the agents- which are autonomous and share similar basic characteristics. The
agents constantly interact with each other based on local information and simple rules. These simple interactions often lead to the emergence of some global structure patterns which cannot be predicted simply from the properties of the individual agents. Agent-based model have certain advantages over traditional analytical models. For example, analytical models often assume homogeneity within the system due to the limitation of mathematical formulations, and the interest of study is the equilibrium state or the average characteristics of a system. In contrast, agent-based models study the transient behaviors of a system before it reaches equilibrium. Agents are not necessarily homogeneous, but differ in their properties or behaviors. This heterogeneity is commonly observed in real systems. Moreover, while analytical models often assume infinite populations, agent-based models take into account finite populations with different population structures, which have been shown to have a profound influence on system dynamics.
Although computer modeling is well-established in the connectionist study of language acquisition, it is a relatively recent, although rapidly burgeoning, development in the study of language origin. According to Nowak (2001), "computer models may adopt different paradigms of language evolution, being a biological or cultural transmission process, or a co-evolving process." Most models study the emergence of one of the subsystems of language, for example phonology, vocabulary, or syntax. Many of these models are agent-based models. For example, Steels (1998) and Ke et al. (2006) studied the emergence of a simple lexicon. These models demonstrate how a set of arbitrary associations between meanings and forms can be established as conventions through imitation and self-organization in a group of agents. While theses models assume the pre-existence of meanings, Steels and Kaplan (2002) presented models where meaning are not prefixed but co-evolve with the meaning-form associations.
There have also been models investigating the emergence of sound systems, such as de Boer (2001), for vowel systems, and Oudeyer (2002) for syllabic structures. Although these models consider only sounds, without the presence of meanings, they can produce results very close to the universal distributions of sound systems found in real languages, which suggest that the assumptions in these models are highly probable. A few models have worked on the emergence of higher-level linguistic structures: Batali (1998), Kirby (1999), and Gong et al. (2005) studied the emergence of compositionality, and Kirby (2002) simulated the emergence of recursive structures. These models are all highly simplified, and the assumptions can be controversial, but they are important initial steps in the area of modeling language origin.
In the agent-based models of language origin, individual language users are the agents. These agents share similar characteristics, for example, articulation and perception of sounds or some general learning mechanisms such as imitation and association, or recurrent pattern extraction. The representation of language in the agents is usually one of two types. One involves neural networks, which are characterized by their distributed nature. The input of the network may be the meaning represented by some grounded features of physical objects such as color, size, and shape, etc., and the output of the corresponding linguistic form or signal. Conversely, the input of the network may be the signal and the output the meaning. The other type of representations, according to Kirby (1999), is symbolic, where meanings and forms are all represented by discrete symbols, such as lexical mappings, or syntactic rules.
In an agent-based model, while agents are assumed to be governed by similar underlying mechanisms, they do not necessarily behave in exactly the same way. For example, they do not necessarily develop exactly the same language. Furthermore, even though they appear to share a language, their internal representations may be different. As Milroy (1987) pointed out, "what the agents learn and how they use their language depend on the histories of their interactions with the environment, which highly depend on their social status and social connections, as evidenced by empirical findings in studies of social networks." However, according to Ke (2006) and Nettle (1999), the factors that cause heterogeneity have not been much explored in the models of language origin, although there have been some attempts in models of language change.
In addition to the consideration of implementing individual agents, Briscoe (2002) stated that it is necessary to move from the study of individual (idealized) language learners and users, endowed with a LAD and acquiring an idiolect, to the study of populations of such generative language learners and users, parsing, learning and generating a set of idiolects constituting the language of a community. The interactions between agents may take place in a random way, that is each time two randomly selected agents interact. Alternatively, agents may interact only with the nearest neighbor or with a number of neighbors within a certain distance (such as models of language change, e.g. Nettle (1999). Gong et al. (2004) is one of the few studies which examined the relationship between language and social structures. It is shown that different communication strategies lead to different social structures: a random interaction strategy results in an almost fully-connected network and a strategy with a preference to a popular agent in a local world results in a more sparse and segregated network
8. CONCLUSION
The origin and evolution of language is one of the major unresolved problems of science, despite a long history of research on the subject. There is at the moment a growing research effort to model and synthesize processes underlying the origins and evolution of language. This paper adopted an emergenist perspective for the study of language origin, which provides a more effective approach to addressing language origin than the natitivist view which has dominated the field for decades. While nativism attempts to explain the origin of language by examining mostly the biological endowment in individuals, emergenism, by contrast, advocates examining the effect of long-term interactions between individual language users. Emergenism concentrates on the emergence of language at the population level. Research on biological explanations for language origin will benefit from this shift, by asking more pertinent questions about the initial conditions for language acquisition and language origin. These initial conditions are unlikely to be the highly abstract, innate mechanisms for syntax proposed by UG theorists, such as c-command or the subjacency principle, and so on, for which the universality in existence and representation are dubious. Instead, low level mechanisms and capacities, such as intention detection, imitation, sequential ability, analogy, and so on, may be more relevant. Although it is still unclear yet, if these abilities are sufficient to account for a fully-fledged language, it is helpful to see what these simple capacities can lead to. While it is hard to examine the long term effects of interactions in empirical studies, computer models provide an effective way of studying the actual
emergent process in a controllable manner, and of examining the effects of variables and parameters.
The emergenist perspective adopted for the study of language origin shares a central idea with the study of language acquisition, which is that unexpected structures come into being spontaneously as a result of long-term interactions between components in the system, and the structures con not be explained simply by examining the individual components. The emergence that language origin and language acquisition are concerned with, however, is at two different levels. Emergence in language acquisition takes place at the level of individual learners, as a result of interaction between innate abilities in learners and their experiences in the environment. In contrast, language origin is emergent across a longer time span at the level of population, as a result of interactions between different individuals in the speech community. Nevertheless, investigations of two levels inform each other. As illustrated in this paper, the model of language makes use of findings from the study of language acquisition. In this way, the model shows how phylogeny can be studied by recapitulating ontogeny. At the same time models of origin raise questions for empirical study of language acquisition. In particular, during computer modeling, as every assumptions has to be made explicit and implementable, specific questions arising from the design of models, such as whether decompositions happen when recurrent patterns are extracted from the input, how homophony is treated by children, and so on, will pose new research topics for psycholinguistics and corpus studies.
Implications for Language Teaching and Learning
What contributions or insights could the study of language origin from an emergenist perspective provide for language Teaching and Learning?
First of all, the study of language origin addresses questions concerning the nature of human language and its defining characteristics. These intriguing questions would lead us to a bigger picture when we study and teach language. From an emergenist perspective, language is dynamic, perpetually evolving and constructed in a piece-meal manner, not only in the individual but also in the population. This will remind us of bearing a balanced view of language between its biological and cultural aspects. Then, we may be more careful not to ascribe the observed regularities in language development too rapidly to learners' shared biological predispositions. We will look more closely at the contributing factors in the learning environment and the learning process.
Secondly, what has been highlighted in the emergenist view for language origin can find parallels in many current thoughts in the field of psycholinguistics. For example, an agent's cognitive apparatus for learning and interaction is made very clear at the beginning of the model; this should find close connections with the studies of cognitive linguistics in first and second language acquisition, as well as the connectionist models which emphasize the use of general cognitive abilities for language learning. Interaction is the crucial source of emergence. In the model discussed in this paper, the agents construct their own language through interactions with others. The input that agents receive therefore determines their language development. This is in line with the various input-based theories of SLA (e.g. Krashen 1985), and the current model can extended to study the relation of input and the regularity of development. The social and cultural
factors play crucial roles in the process of individual's learning, as has been recognized in the study of SLA. Moreover, agent's language development in the model is similar to the interlanguage development studied in SLA, which is viewed as a dynamic construction process in its own right, instead of an unimportant intermediate transition toward a static target. As learning is a self-constructing process, it is very important to raise learners' awareness and direct their attention to patterns in the learning input and also to their own errors. Tomassello and Heron (1988) have suggested a 'garden path' technique to lead learners to make errors and then learn from them. For example, to learn past tense in English, learners are first given the rule, which naturally results in overgeneralization, such as 'eated' for 'ate'. Once they make an error, and only after they have actually made an error, learners receive feedback on their errors. It is shown that this method is more effective than telling learners in advance about exceptions to a rule.
Thirdly, the highly interdisciplinary nature in the study of language origin may provide language teaching and learning with insights into exploring new research methodologies and cross-discipline collaborations. Computer modeling may be one productive area to experiment.
The agent-based model highlights two important features of emergent phenomena: heterogeneity and nonlinearity. As we have seen from the model, even though the population as a whole can achieve a high mutual understanding between individuals' languages, that is the idiolects, differ from each other from the very beginning. In real life, children exhibit different growing patterns in their language development. These individual differences are even more prevalent in SLA, not only in their observable linguistic behaviors in the process of learning, but also in cognitive mechanisms underlying language aptitude, motivation, learning styles, and so on. Though the issue of 'learner variety' has long been recognized, there is not enough actual research and teaching practice yet. It is necessary to recognize heterogeneity in learners in every stage of learning, and provide individually based feedback as much as possible. It is important to highlight the fact that there is no single standard language to learn. Instead, language exists as a large variety of idiolects dependent on different genres, speech styles, social classes, etc. Therefore, it is important to raise students' awareness of not only the regularity, but also variation, and instability in actual language use.
Another distinctive feature of emergent systems is the existence of nonlinearity and phase transition. The dynamics of system does not proceed in a linear way. Sometimes, the system may go through sharp transitions with abrupt changes, even when there is no abrupt change in either the external input to the system or the internal parameters of the system. In the process of SLA, there are many such sharp transitions. In order to be able to observe these transitions, we have to zoom in on the right time period and scrutinize the intermediate changes within that window. Otherwise, when this short time frame is missed, one observes the two plateau stages before and after the transition, and misses the rich characteristics in the transition period.
Non-linearity has two significant implications: (i) in order to understand how learning progress, we have to pay special attention to capturing such abrupt transitions and find out if there are particular conditions or prompts that trigger such transitions; (ii) we will expect plateau period, and provide continuing support to learners even though at times there seem to be no significant progress.
References
Akhtar, Nameera. 1999. "Acquiring Basic Word Order: Evidence For Data-Driven Learning Of Syntactic Structure." Journal of Child Language 26 (2): 339-356. https://doi.org/10.1017/s030500099900375x.
Batali, John. 1998. "Computational Simulations of the Emergence of Grammar." In Approaches to the Evolution of Language, edited by James R. Hurford, M. Studdert-Kennedy, and Chris Knight, 405-426. Cambridge: Cambridge University Press.
Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, Mass: MIT Press.
De Boer, Bart. 2000. "Emergence of sound systems through self-organization." In The evolutionary emergence of language: social function and origins of linguistic form, edited by Chris Knight, Michael Studdert-Kennedy, and James R. Hurford, 177-198. Cambridge, MA: Cambridge University Press.
Ellis, Nick C. 1998. "Emergentism, Connectionism and Language Learning." Language Learning 48 (4): 631-64. https://doi.org/10.1111/0023-8333.00063.
Hauser, Marc D., Noam Chomsky, and W. Tecumseh Fitch. 2002. "The Faculty of Language: What Is It, Who Has It, and How Did It Evolve?" Science 298 (5598): 1569-1579. https://doi.org/10.1126/science.298.5598.1569.
Ke, J., and J. H. Holland. 2006. "Language Origin from an Emergentist Perspective." Applied Linguistics 27 (4): 691-716. https://doi.org/10.1093/applin/aml033.
Kirby, Simon. 1999. "The Role of I-Language in Diachronic Adaptation." Zeitschrift Für Sprachwissenschaft 18 (2): 222-225. https://doi.org/10.1515/zfsw.1999.18.2.222.
Kirby, Simon. 2002. "Natural Language from Artificial Life." Artificial Life 8 (2): 185-215. https:// doi.org/10.1162/106454602320184248.
Krashen, Stephen D. 1985. The input hypothesis: issues and implications. London; New York: Longman.
Larsen-Freeman, D. 1997. "Chaos/Complexity Science and Second Language Acquisition." Applied Linguistics 18 (2): 141-65. https://doi.org/10.1093/applin/18.2.141.
Lee, Namhee, and John H. Schumann. 2003. The evolution of language and the symbolosphere as complex adaptive systems. Paper presented at the Conference of the American Association for Applied Linguistics [Arlington VA, March 22-25].
Macnamara, John, Ray Jackendoff, Paul Bloom, and Karen Wynn, eds. Language, Logic, and Concepts: Essays in Memory of John Macnamara. Cambridge, Mass: MIT Press, 1999.
MacWhinney, Brian, and William O'Grady, eds.. 1999. The Emergence of Language. Mahwah, NJ: Lawrence Erlbaum Associates.
Milroy, Lesley. 1987. Language and Social Networks, 2nd ed. Oxford: Basil Blackwell.
Nettle, Daniel. 1999. Linguistic diversity. Oxford: Oxford University Press.
Oudeyer, Pierre-Yves. 2002. "Novel Useful Features and Algorithms for the Recognition of Emotions in Human Speech." Proceedings. of the 1st International Conference on Speech Prosody.
Pinker, Steven, and Paul Bloom. 1990. "Natural Language And Natural Selection." Behavioral and Brain Sciences 13 (4): 707-727. https://doi.org/10.1017/s0140525x00081061.
Rvachew, Susan, and Michele Nowak. 2001. "The Effect of Target-Selection Strategy on Phonological Learning." Journal of Speech, Language, and Hearing Research 44 (3): 610623. https://doi.org/10.1044/1092-4388(2001/050).
Scovel, Thomas. 1998. Psycholinguistics. Oxford Introduction to Language Study Series. Oxford: Oxford University Press.
Steels, Luc. 1995. "A Self-Organizing Spatial Vocabulary." Artificial Life 2 (3): 319-332. https://doi.org/10.1162/artl.1995.2.3.319.
Steels, Luc. 1998. "The Origins Of Syntax In Visually Grounded Robotic Agents." Artificial Intelligence 103 (1-2): 133-156. https://doi.org/10.1016/s0004-3702(98)00066-6.
Steels, Luc. 1997. "The Synthetic Modeling of Language Origins." Evolution of Communication 1 (1): 1-34. https://doi.org/10.1075/eoc.L1.02ste.
Steels, Luc, and Frederic Kaplan. 2002. "Bootstrapping grounded word semantics." In Linguistic evolution through language acquisition: formal and computational models, edited by E. J. Briscoe. Cambridge; New York: Cambridge University Press.
Tomasello, Michael. 2003. Constructing a language: a usage-based theory of language acquisition. Cambridge, Mass: Harvard University Press.
Tomasello, Michael, and Carol Herron. 1988. "Down the Garden Path: Inducing and Correcting Overgeneralization Errors in the Foreign Language Classroom." Applied Psycholinguistics 9 (3): 237-246. https://doi.org/10.1017/s0142716400007827.
Tomasello, Michael, Josep Call, and Andrea Gluckman. 1997. "Comprehension of Novel Communicative Signs by Apes and Human Children." Child Development 68 (6): 1067. https://doi.org/10.2307/1132292.
Williams, Marion, and Robert L. Burden. 1997. Psychology for language teachers: a social constructivist approach. Cambridge language teaching library. Cambridge; New York: Cambridge University Press.
Declaration of conflicting interests
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of
this article.
Funding
The author received no financial support for the research, authorship and/or publication of this article.
Author information
Ali Rahimi is an Assistant Professor at Kashan University, Iran.
Hamzeh Haghighi is an MA Student at Kashan University, Iran.
Copyrights
Copyright for this article is retained by the author, with publication rights granted to the journal.
