Can we consider language a complex adaptive system?
If so, how can we model it, and how can we test a model like this?
We tried a model considering it as a system composed of four subsystems/events (Lexicon, Discourse, Semantics and Grammar) in a two-mode network (we proposed eleven hypotheses for this). Speakers are the elements interacting through the events. Comments, critics, and suggestions are welcomed for discussion, in special considering possibilities of data to test the model.
Article Modeling the Complexity of Language As a Two-Mode Network
Philadelphia, PA
Dear Filho & readers,
I suspect this question and discussion requires some more or less technical definition of "complex adaptive system." Lacking something of this sort, I suspect possible answers may drift about depending on how the phrase is understood on a more intuitive basis.
I am sympathetic to the question above, too, about the distinction involved between lexicon and semantics--in particular. What would "lexicon" mean, if we can have the same lexicon and a different semantics? Also, one may wonder how the specification of the semantics of a system will contribute (or fail to contribute?) to the differentiation of languages. Can we have the same language, in system 1 and system 2 if they differ in semantics?
On the other hand, I'm inclined to understand "grammar" as a matter of syntax.
Give "systems" frequently "adapting" one might suspect, in the end, that this will yield a very fine-tuned differentiation of "languages."
H.G.Callaway
It is obvious that language meets the features of a complex adaptive system, but I doubt that it can be modeled in such a simplistic way as having only four discrete components. It's not clear how you would quantify a system like "semantics", or even why you chose these four systems (how is "semantics" different from "lexicon" or "grammar"?).
Nowadays linguists talk a lot about interfaces, i.e., the traditionally established components of grammar are not clear-cut compartments for a specific aspect of language to be stored. For example, in corpus linguistic study we begin to realize the importance of lexical bundles, or the frequently occuring word sequences, this phenomenon essentially stridles lexicon and syntax, or it is a lexico-grammatical phenomenon. Human language is a very complex system, constantly adapting to changes both from within and from without, so personally I doubt any model can sufficiently explain this complex system, the same is true for any existing models of grammar (in the sense of linguistic theory).
Hello,
You are right to say that language is a very complex adaptive system. A close look at first language acquisition by infants reveals that language development is entwined with cognitive and social dimensions whose labyrinthine nature makes computational modelling a very daunting task. Computational models such as the one you describe are based on linear models of linguistics which operate based on competence algorithms. However, language used for communication involves performance factors and encompass parallel mechanisms with non-linear functioning which makes it very difficult to be modeled. All in all, situational, sociolinguistic and pragmatic factors have unpredictable behaviors which pose a great challenge to modelling.
Best regards,
R. Biria
this link is useful
cnl.psych.cornell.edu/pubs/2009-LACAS-pos-LL.pdf
regards
Philadelphia, PA
Dear Filho & readers,
I suspect this question and discussion requires some more or less technical definition of "complex adaptive system." Lacking something of this sort, I suspect possible answers may drift about depending on how the phrase is understood on a more intuitive basis.
I am sympathetic to the question above, too, about the distinction involved between lexicon and semantics--in particular. What would "lexicon" mean, if we can have the same lexicon and a different semantics? Also, one may wonder how the specification of the semantics of a system will contribute (or fail to contribute?) to the differentiation of languages. Can we have the same language, in system 1 and system 2 if they differ in semantics?
On the other hand, I'm inclined to understand "grammar" as a matter of syntax.
Give "systems" frequently "adapting" one might suspect, in the end, that this will yield a very fine-tuned differentiation of "languages."
H.G.Callaway
yes of course. Languaje is a system that consists of the development, acquisition, maintenance and use of complex systems of communication, particularly the human ability to do so; and a language is any specific example of such a system. Human language has the properties of productivity and displacement, and relies entirely on social convention and learning.
Dear H.G.Callaway.
Thanks for your very helpful comments. First off all I think you touched a crucial point when talking about the definition of “complex adaptive systems”, or simply CAS. It appears that most understand “systems” and “complex” in varied forms, and this will direct very different approaches. And the answer of Hassan Nima (thanks!) helps because the link suggested (Language is a complex adaptive system) came with definitions, like the key features of a CAS: multiple agents interacting with one another, adaptive, dynamical, competition, emergence, patterns of experience, social interaction, cognitive processes.
In my view a precise definition that implies the key features could be: “complex system is an ensemble of many elements which are interacting in a disordered way, resulting in robust organization and memory” (from Ladyman, Lambert, and Wiesner at “What is a complex system?”).
For us, it is not possible to derive lexical, discursive, semantic and grammatical categories one from the other, for they live together and actuate in self-dynamic systems. There is no uni-directionality in their relationships since this principle is helpful only to explain changes inside phonological, morphological, syntactical and semantic systems.
Our tentative is to create a simple model (not a simplistic) just to simulate possibilities of interaction and development. In fact, how to do a simulation to answer your question: “Can we have the same language, in system 1 and system 2 if they differ in semantics?”
Thanks again, Callaway!
Mauro
Dear Filho,
It appears to me that what you call 'language' in your article linked in your question is language use in communication. As you show, this is follows from the first conjunct of hypothesis 3. “Language has exactly four distinct events and a minimum of two actors (speakers)..“ if the four components are “lexicon, semantics, discourse, and grammar”. I fail to detect any argument in favour of the claim that discourse (hence communication) is a relevant event for language. To be sure, we use language to communicate, but apart from common sense there is not much reason to believe that communication is essential for language (and common sense is irrelevant for scientific inquiry). In fact, the hypothesis seems false, since an individual can talk to herself in internal speech and it is possible that she is the last survivor that understands her language, according to you the language she uses does not exist, which is contradictory.
If you delete discourse, then you are left with the notions of the lexicon, the syntax (“grammar”) and semantics. The study of language could then focus on the individual, as it is standard in biolinguistics. That is to say, language is an internal biological property of an individual (Chomsky 2007, p. 1). This approach has the advantage that you get immediately a clear criterion for the evaluation of the model: is the model isomorph to the actual processes? In other words: is your model an adequate representation of the physically realized processes in the mind/brain of an individual (for the three components of the model that must be physically realized in the mind/brain of every competent speaker)? Admittedly, it is difficult to test this directly since the physical processes underlying language are badly understood, although there is interesting work by Piattelli-Palmarini and Vitiello. They argue that there is an isomorphism between linguistic minimalism and quantum field theory, suggesting that the underlying processes may be found on the quantum level (Piattelli-Palmarini and Vitiello 2017).
This might leave only the familiar indirect methods of theoretical linguistics to test the model: descriptive adequacy, explanatory adequacy and what one may call “beyond explanatory adequacy” (Chomsky 2007, p. 2). If language is an internal biological property, then your model must fulfil the adequacy conditions for the three components (which seems difficult enough and I suspect hopeless if you add something as complex as human interaction i.e. “discourse”).
Best,
Sven Beecken
Noam Chomsky; Biolinguistic Explorations: Design, Development, Evolution; International Journal of Philosophical Studies Vol. 15(1), 1–21; 2007
M Piattelli-Palmarini and G Vitiello; Quantum field theory and the linguistic Minimalist Program: a remarkable isomorphism; IOP Conf. Series: Journal of Physics: Conf. Series 1234567890
880 (2017)
Philadelphia, PA
Dear Beecken & readers,
That discourse and communication are relevant to language would seem to follow simply from the fact that discourse and communication are relevant as evidence. However abstract, "biological" or even mathematical, e.g., the concept of language may be in any given approach, it would seem that discourse and communication belong to the relevant facts on the ground. The point seems also to be suggested by the primacy usually attached to verbal language as contrasted with written language. But even written communication and discourse can be understood as relevant evidence for understanding and accounting for the phenomenon of language.
What comes to mind, in the first place is the relevance of actual usage of words to empirical lexicography. This would seem to be central to empirical semantics. But we understand usage partly in the context of discourse and communicative practices--and even related activities. In contrast, your biological (or psychological?) concept seems very abstract --something standing in need to a great deal of ground-level empirical evidence. I don't mean to rule it out, of course, but to emphasize the question of how one might support it --or any alternative.
H.G. Callaway
---you wrote---
I fail to detect any argument in favour of the claim that discourse (hence communication) is a relevant event for language. To be sure, we use language to communicate, but apart from common sense there is not much reason to believe that communication is essential for language (and common sense is irrelevant for scientific inquiry). In fact, the hypothesis seems false, since an individual can talk to herself in internal speech and it is possible that she is the last survivor that understands her language, according to you the language she uses does not exist, which is contradictory.
Dear Callaway,
If one considers language use in communicative situations as indirect evidence for linguistic competence, then there is no need to include communication into the range of phenomena that one wishes to explain. The distinction between competence and performance where performance is taken as indirect evidence for competence can be traced back (at least) to Chomsky's Aspects of the Theory of Syntax. Of course, one can choose to include communication (and whatever else one wants) into the range of phenomena to be explained, but this is mere stipulation.
On the other hand, one can give arguments for the claim that language must consist at least of the lexicon, syntax and semantics. Recent work by Chomsky and others strongly suggests that language consists at its core of a recursive procedure that combines the elements selected from the lexicon into expressions. These expressions are interpreted at two interfaces, the interface for semantic interpretation and the interface for phonetic interpretation i.e. externalization. The archaeological record suggests that language has emerged in the last 50 000 to 200 000 years, the system for externalization exists far longer, this indicates that there are in fact two systems, one consists of the recursive procedure and the semantic interface and the other of a system for externalization. This might indicate that language is at its core a system for thought and externalization is a secondary aspect, which entails a fortiori that communication is a tertiary aspect of language.
This is necessarily a rather brief and weak exposition of what I had in mind when I wrote that language is an internal biological property. I don't want to quarrel about terminology, I'm taking this directly from Chomsky and he regards the study of the language faculty (“syntax”) as the study of the genetically determined factor influencing the growth and development of linguistic competence.*
I do agree that this theory requires substantial evidence, but if it can be established (and Chomsky and others seem increasingly confident that it can) it would strongly suggest that communication and language are two very different things that should be studied separately in order to gain understanding.
A much better exposition then mine of the theory and substantiations of many of the claims made above can be found in the rather short and concise paper How could Language have Evolved? by Johan Bolhuis, Ian Tattersall, Noam Chomsky and Robert Berwick, PloS Biol 12(8), 2014.
Best,
Sven Beecken
* See, for example, Chomsky's Three Factors in Language Design; Linguistic Inquiry, Volume 36, Number 1, Winter 2005.
Dear H.G. and Sven,
My appreciation for your insightful discussion of the language model proposed by Filho et al. I have two comments proceeding from my own experience of consciousness and observation of the world.
First, Sven's view that lexicon, semantics, and grammar comprise sufficient primary events to define the occurrence of language matches my experience. I experience discourse, whether internal (i.e. within my own mind) or apparently external (i.e. supposing that "others" exist independently of my own mind) as a derivative event that is by no means limited to "language" in the ordinary sense of that word. From my perspective, discourse is a product of language plus non-language semiotic events. Even though discourse observably possesses the power recursively to manipulate its constituent language-events, I do not experience that it is a component of language. Discourse is like the activity of an enzyme; language is like DNA. An enzyme's activity may include manipulating DNA, including even the very DNA which defines its own amino acid sequence. But the fact that an enzyme manipulates the components of DNA obviously does not mean that the enzyme itself is a component of DNA.
Secondly, H.G.'s question: What would "lexicon" mean, if we can have the same lexicon and a different semantics? provides a good illustration of the hierarchically superior position of discourse in relation to language. Situations like this appear to be quite frequent occurrences in discourse. For instance, U.S. constitutional law is a discourse in which the text (lexicon and grammar) remains quite stable but its meaning (semantics) shifts dramatically in response to world events. Taking an example of a different kind: also comparatively simple language systems, such as chess, occasionally experience such meta-events. Introduction of the en passant rule was one instance of this. Following this event, the chess lexicon (pieces and board) remained stable but the associated semantics were altered (in this case due to introduction of a new rule of grammar - i.e. a new way to capture a fourth-rank pawn under certain conditions).
I hope these comments are of some use in advancing the work of Filho et al. and usefully contribute to discussion of the associated concepts.
Kind regards,
Michael Lusk
Philadelphia, PA
Dear Beecken & readers,
I have no objection to the customary distinction between competence and performance in Chomsky and others. I tend to think of this, however, as concerns semantics, as parallel to the distinction between meaning and usage --regarding actual usage as the evidence of meaning in empirical lexicography. (Think of the writers of dictionaries collecting vast numbers of examples of usage--as in the famous history of the OED.)
Usage, however, is found and examined or understood in particular contexts, which leads one from words to sentences (statements if you prefer), and from sentences to larger texts and the situational contexts of use and activities--including exchanges in discourse. All of this may reasonably enter into the evidence for ascriptions of meanings to lexical items.
In consequence, though it may be that the study of discourse and communication has its own goals and some separate integrity as a discipline, it does not follow that such studies could have no valid influence upon empirical semantics and accounts of language. In a similar way, though physics, say, is distinct from chemistry or biology, still the one science may reasonably exercise some constraint on the development or problems of the others.
H.G. Callaway
---you wrote---
I do agree that this theory requires substantial evidence, but if it can be established (and Chomsky and others seem increasingly confident that it can) it would strongly suggest that communication and language are two very different things that should be studied separately in order to gain understanding.
Dear Callaway, Michael, Filho and readers,
I must apologize for the unclear way to express myself. I did not mean to suggest that evidence from usage could not have influence on the study of language or semantics, on the contrary, it seems that this is at the moment the major source of evidence. My original criticism of the model proposed by Filho and Queriquelli is that communication is far too complicated to have any hope of achieving even descriptive adequacy. Consider Callaway's example, the construction of expressions and their use in communication from the perspective of Chomsky's theory: The internal computational procedure selects the word-like elements from the internal lexicon and constructs the expressions. These expressions are then interpreted at the semantic interface and fed into other systems for thought, planing, action and so on (I'm not suggesting that we have much insight into how these systems work, only that its a plausible conjecture that they must exist). If we then consider the use of these expressions in discourse, we have to consider all the influencing factors, from the fact that the individual acts over the beliefs, goals, wishes all the way to body language, closing and whatever else influences communication (after all, one can change the meaning of an utterance by raising an eyebrow).
It is worth pointing out that the elements that enter into computation have properties that go far beyond anything that even the most comprehensive dictionary has to offer. In this sense, the internal lexicon has not much to do with a lexicon in the sense of the OED. As I said, I do agree that usage gives indirect evidence for the study of these elements, the interesting point is that we can study this even if the external world would not exist. Just as a neuroscientist may study insect navigation in a laboratory where the sun is replaced by an artificial light (all that matters is sensory stimulation). In that sense, it seems that what we are actually studding when we study usage is something that primarily happens internally within the mind/brain of an individual (or individuals). Chomsky has argued this for a long time.
I think it is interesting that this view closely matches Michael's description of the phenomena that we can have access to via introspection. The real question is: what are the underlying mechanisms that bring fourth these phenomena? This is the reason why I am sceptical with regard to the range of phenomena that Filho's and Queriquelli's model is supposed to capture. Even if descriptive adequacy could be achieved, it would still be the case that there seem to be many mechanisms involved, as indicated above. The more mechanisms are involved in bringing forth the observed phenomena, the more difficult it gets to actually understand these mechanisms. That is why I suggested to separate the range of phenomena to be explained.
Best,
Sven Beecken
Dear Michael Lusk, H.Calloway and Sven Beecken.
When we were thinking about a model to simulate language, we were inclined to use only three events (Lexicon, Grammar and Semantics), considering that Discourse would be the signalling process of the Complex Adaptive System – CAS. In this case, we would have three events on one side and the Actors (speakers) on the other. However, we understood that discourse is a fundamental component of the system because, in fact, this is the event responsible for the “adaptive” word at “CAS”, that is, with no discourse, no evolution. For this reason, to have not a frozen language, not a “static language” (does it exist?), discourse is one of the four that affects, all the time, any other event: grammar/lexicon/semantics (anyway, in our model, the four events are in the same level – there is not a superior “event”).
Yet, for this reason, language needs at least two speakers. If there is just one, like Sven described, there´s still a language, but not any more as a CAS. The first three hypotheses of our tentative model presents just this. Another hypothesis (number 9) says that the links (communication) can have different bandwidths, in a way that the influence and intensity from one event to the others can vary all the time but are always present.
I agree it's a difficult computation, so we're trying the simplest rules to deal with and then seeing what the interacting elements could show us. Surely this discussion is helping a lot the advancement of our work - to refine our model - or even discard it. Thanks again for your contributions!
Mauro Faccioni
----the hypotheses cited:
Hypothesis 1. Language is a complex system comprising a two-mode network, where the mode of events consists of four subsystems (Lexicon, Semantics, Discourse and Grammar) and the mode of actors is made up of the speakers of this language.
Hypothesis 2. Language is a two-mode network completely connected and dense.
Hypothesis 3. Language has exactly four distinct events and a minimum of two actors (speakers).
Hypothesis 9. Language is a complex system wherein the links could have variable values, or widths.
Philadelphia, PA
Dear Faccioni & readers,
I think we have made a good case for the relevance of discourse to the understanding of the concept of language. I would say , too, though that one must be careful both with the concept of "discourse" and the concept of "adaptation." Not every adaptation is perhaps a happy or felicitous adaptation; and speakers are not always completely neutral regarding alternatives found in "discourse." Nor need they be.
I imagine, to take an extreme, a "discourse." say recognizably in English, between two of these computerized, talking and answering mechanisms that the large technology firms seem to be selling to everyone. One might conceivably overload a corpus with the results of their back and forth replies (say between two of them); and we might then still have good reason to doubt that this "discourse" was of interest or fully relevant to the characterization of any "language." In a sense, I rely on Wittgenstein, who wrote that to understand a language is to understand a form of life. "Adaptation" to the computerized "discourse" might just take us nowhere of interest. More generally, I am registering some skepticism concerning the notion of very general and formalistic conceptions of language. It is not that formalistic approaches never have any use, but one must be careful about rushing into it.
I noticed, earlier, some possible ambiguity in your discussions of the concept of "complex adaptive system." At times the notion sounded almost purely mathematical--except, perhaps, for the concept of an "agent." (An agent, I take it is someone or something that brings about activities, or specifically linguistic activities.)
In addition, it is well remarked that we learn language, or our first language at least, in discourse and human interaction. (Notice that our first babblings are left out of the corpus.) But I do not see that this origin quite literally and flat-footedly belongs to the very concept of language. More to the point is that there is no private language, in the sense of one that could not be used in communication. Whether a given language is used in communication, at any given time, is a different question.
Many thanks for your comments and additions.
H.G. Callaway
(BTW: notice the 3 "a's")
Dear Mauro and all participants,
Condradulation for the paper! At last - something of reasonable which is not deep learning.
Interresning point in the paper - language is a process.
Mauro, I didn't enter into deep details when reading the paper. However, the work looks at language as a coding system designed to exchange.
To exchange what?
Well, I start with the skepticism (this is Not critics):
The issue wich is exchanged, accorging to me, is meaning. The stuff which is a process, according to me, is not the coding system, but the Meaning.
One and the same statement can have different meanings - not becouse of the context, not becouse of the prosody, not becouse of the listener etc.... but becouse of the state of the listener. The substance wich is usally labeled Semantcs is created inside the head. Meaning is something of generated - by the human.
Codding systems, it seems, can evolve spontaneousely. I do a copy-paste here, esxuses : " Studies have been undertaken in the domain of robotics, aiming to reproduce the emergence of symbolic systems for communication. A study (Grouchy et al, 2016 https://www.nature.com/articles/srep34615) demonstrates that in 99% of all experiments in a simulated world, 9% of the robots developed communication schemes. The constraint in this world was that robots’ ability to reproduce was motivated by the need to find a mate. "
I think that we will succed to model the coding system if we knew how is made the substance which is exchanged.
My words:
"More generally, I am registering some skepticism concerning the notion of very general and formalistic conceptions of language.""
but H.G. Callaway wrote them before me...
Regards,
Velina
Dear Velina,
thanks a lot for your comments ! I´m glad with your scepticism and critics, with them I think I can reason a bit more about our model and its problems/failures.
I agree with you about "meaning" as a stuff that is exchanged. But the meaning is a question of each individuality (each actor in our model). Them the meaning is translated to a physical language and then the language evolves as the comprehension of actors process the words, sentences, etc.
The model proposes a network with actors interacting. Each actor reacts to some message coming and reply depending on how he/she interprets the message and the “internal meaning” reacts to these stuff. Well, our model do not (at least for a while) have a “reaction algorithm” for the actors. That is, if a message reachs an actor, how will be the reply? I don´t know, for sure it depends on the “meaning”. I suppose that the meaning comes from a way that we process internally the metaphors, differently of the computers that reply depending on a growing storage of possible responses.
Well, these are suppositions, anyway.
Thanks Velina, and I will be very grateful if you have any other suggestions/comments!
Mauro
A very interesting reference could be the book by Ricard Solé "Phase transitions", where he dedicates a whole section to critical phase transitions in models of language evolution (e.g. the naming game).
Another great paper about language as a complex adaptive system is the work by Baxter and colleagues about the complex, non-deterministic emergence of the New Zealand English from a variety of historical happenings and demographical processes:
https://www.cambridge.org/core/journals/language-variation-and-change/article/modeling-language-change-an-evaluation-of-trudgills-theory-of-the-emergence-of-new-zealand-english/0264FEC709F3CCFDC2CBCED0F3408230
More recently, another great paper about language as a complex system with plenty of heavy-tail and "power-laws" is the work by Ferrer i Cancho and Vitevitch, who showed how Zipf's power-law rule of meaning distribution can be elegantly obtained from Zipf's power-law of word frequency:
Article The origins of Zipf's meaning-frequency law
Dear Vasilii A. Gromov and Massimo Stella,
thanks for your references and suggestions.
Dear Luciano Costa,
thanks for your answer and the viewing of language as not completely comprehensive in its role for communication. Definitely, I agree and I suppose that is not possible to measure how much of a word is understood by a speaker, but we could add a probability on this. Imagining the group of speakers doing their dialogues, talks will be the connections (through the events proposed in the model). A weight (the understanding probability) is add to each link. On the other hand, maybe, putting this weight within the actor to shape his/her answer time by time.
Well, good ideas to implement. Thanks!
It seems to me that it is. But, it is a human-driven CAS just as I said in a short comment: Article Language as a human-driven complex adaptive system