I am not a language researcher, but borrowing from research in other domains, it seems that your finding—that a richer exchange/connectivity accelerates convergence—is unsurprising. is this not the case in language development/research?

To what extent does evidence indicate that convergence can be productively changed by overt means (schools, for example), in contrast to the more naturalistic convergence that you have shown for the families and communities?

How might you expect language to evolve with the increased connectivity provided by texting and the internet compared to traditional forms of spoken and written communication?

I have two related questions: (1) How do your results and proposed processes fit with language changes previously described in the literature regarding the shift from a pidgin to a creole language? (2) How does the age of the language learner affect the processes you study?

Wow, this is fascinating. I’m really interested in evolutionary processes of symbolic emergence and cultural change. One of the benefits of an ABM is that you get to test scenarios that you cannot measure directly, as well as the group-level outcomes of individual level processes or measurements. To me it is not surprising that more connections creates a faster symbolic convergence. Is it faster per network link? I doubt it.

My question is about language emergence and diversification, and is related to Kristopher’s question. Social identity also emerges dynamically from individual interactions over time in a population, but social identities also tend to diverge over time in a large population. Because social identity restricts communication flow, language follows the same patterns. Thus social identity is key to language stability, but so is limited interconnections with other social groups. Could you speculate on the relative influence of social identity and increasing interconnection on language diversity over the long term?

Also, Russell, I wonder if you have considered any ABM fitting approaches? Measuring individual behavioral variables and then fitting ABM network parameters, say, to the individual outcomes. This can be a very powerful inference engine. NetLogo’s BehaviorSearch does this sort of thing. Best, Tim

Very interesting and valuable methodology. Raises lots of questions, of course. Maybe most important (in my mind) is the role of the lexicon within the language as a whole. Once you’ve got even a rudimentary lexicon, then the process of converging on a new entry seems a more constrained problem — but how does this model relate to the “convergence” on syntax (say)? (What’s your position on Universal Grammar?)

Another question is the phonology (or in this case the gestural components) — the example in your video (’milking") is a prime example of an “onomatopoetic” coinage — the lexicalization constitutes in this case a process of creating and refining a representation of something with very clearly visible/comparable characters. So not yet a model for the “emergence of language,” as I understand it, but lexicons are themselves rather complex and fascinating phenomena. Very promising work!

Thanks for your questions and kind words, Brian! Yes, as the lexicon solidifies, a phonology (system governing the combination of forms to create words), will emerge, and that will certainly constrain the creation of additional words. As for how this model relates to convergence of syntax, my coauthor Charles Yang actually first used a very related model for children’s acquisition of syntax. In that model, when a child hears a sentence, they probabilistically choose a particular grammar (say, a grammar that allows subject-dropping vs one that does not) with which to parse the sentence. They then try to parse the sentence with that grammar, and if they are successful, they increase the probability of selecting that grammar in the future. So, something similar to that might be applied to emergence (rather than acquisition) of syntax! As for Universal Grammar, I remain somewhat agnostic, and our work actually skirts the matter somewhat. That is, we can’t really say whether the lexicon convergence that we see here relies on some language-specific cognitive machinery.

Yes, most of the signs we observe are quite iconic (their form resembles their meaning)! This is actually quite common in sign languages’ early stages. Over time, however, as the signs shift to fit in the emerging phonology (for signed languages: the handshapes, locations, movements, palm orientations, and sometimes facial expressions allowed in the language, as well as the constraints governing their combination, e.g. that a non-dominant hand must mirror the dominant hand or stay still), and as they reduce in form to become easier to produce, they tend to lose their iconicity. As an example, the American Sign Language sign for FEEL used to be produced over the heart, but now it is produced over the center of the chest (it has centralized — a common historical change in signed languages). So something like this could happen in homesign, and very likely is happening/did happen in Nicaraguan Sign Language. Conventionalization is a key part of this process — when everyone recognizes the sign, they no longer need iconicity to understand it, and the form of the sign can shift a bit.

But actually, I would say that iconicity in itself is not an un-linguistic property. Making a sign be iconic does not in itself reduce its combinatorial capacity, or any other deeper linguistic qualities, in any considerable way. Iconicity is only short-lived in the language if, as I mentioned, its form does not fit with the rest of the phonology. Otherwise, it can stick around, as it does in many signed and spoken language (Japanese, for example, has an extensive sound symbolism system: http://en.wikipedia.org/wiki/Japanese_sound_sym...).

Very creative video, I love how you have the videos on the whiteboard.

This is really interesting. I study humpback whales and one particularly mind-blowing area of research is how whales sing the same song during the breeding season and how this song evolves within or between seasons. During fall in Alaska, other researchers tell me we can hear an early less refined form of the song but by the time whales get to Hawaii its fully developed. Without knowing much about this field, I wonder whether your model could predict this convergence.

I have a question about the computer simulations you ran. It seems to me that there’s a confound between number of interactions occurring in a fixed period of time, and the number of dyads that are interacting during that period of time. Have you looked at the number of pairwise exchanges required to reach convergence, as well the number of time steps required?

Really ingenuous and refreshing video concept! Exciting emphasis on the larger push you are making at the end of the video- it really is super important that we start using empirical models to start evaluating natural datasets. Great project!