Minimal Requirements for the Emergence of Learned Signaling

Predictability and Variation in Language Are Differentially Affected by Learning and Production

General principles of human cognition can help to explain why languages are more likely to have certain characteristics than others: structures that are difficult to process or produce will tend to be lost over time. One aspect of cognition that is implicated in language use is working memory-the component of short-term memory used for temporary storage and manipulation of information. In this study, we consider the relationship between working memory and regularization of linguistic variation. Regularization is a well-documented process whereby languages become less variable (on some dimension) over time. This process has been argued to be driven by the behavior of individual language users, but the specific mechanism is not agreed upon. Here, we use an artificial language learning experiment to investigate whether limitations in working memory during either language learning or language production drive regularization behavior. We find that taxing working memory during production results in the loss of all types of variation, but the process by which random variation becomes more predictable is better explained by learning biases. A computational model offers a potential explanation for the production effect using a simple self-priming mechanism.

Communicative Efficiency or Iconic Learning: Do Acquisition and Communicative Pressures Interact to Shape Colour- Naming Systems?

Language evolution is driven by pressures for simplicity and informativity; however, the timescale on which these pressures operate is debated. Over several generations, learners' biases for simple and informative systems can guide language evolution. Over repeated instances of dyadic communication, the principle of least effort dictates that speakers should bias systems towards simplicity and listeners towards informativity, similarly guiding language evolution. At the same time, it has been argued that learners only provide a bias for simplicity and, thus, language users must provide a bias for informativity. To what extent do languages evolve during acquisition versus use? We address this question by formally defining and investigating the communicative efficiency of acquisition trajectories. We illustrate our approach using colour-naming systems, replicating the communicative efficiency model of Zaslavsky, Kemp, Regier & Tishby (2018, PNAS) and the acquisition model of Beekhuizen & Stevenson (2018, Cogn. Sci.). We find that to the extent that language is iconic, learning alone is sufficient to shape language evolution. Regarding colour-naming systems specifically, we find that incorporating learning biases into communicative efficiency accounts might explain how speakers and listeners trade off communicative effort.

Instantaneous systems of communicative conventions through virtual bargaining

People can instantaneously create novel conventions that link individual communicative signals to meanings, both in experiments and everyday communication. Yet a basic principle of natural communication is that the meaning of a signal typically contrasts with the meanings of alternative signals that were available but not chosen. That is, communicative conventions typically form a system, rather than consisting of isolated signal-meaning pairs. Accordingly, creating a novel convention linking a specific signal and meaning seems to require creating a system of conventions linking possible signals to possible meanings, of which the signal-meaning pair to be communicated is merely a sub-case. If so, people will not link signals and meanings in isolation; signal-meaning pairings will depend on alternative signals and meanings. We outline and address theoretical challenges concerning how instantaneous conventions can be formed, building on prior work on "virtual bargaining," in which people simulate the results of a process of negotiation concerning which convention, or system of conventions, to choose. Moreover, we demonstrate empirically that instantaneous systems of conventions can flexibly be created in a 'minimal' experimental paradigm. Experimental evidence from 158 people playing a novel signaling game shows that modifying both the set of signals, and the set of meanings, can indeed systematically modify the signal-meaning mappings that people may instantaneously construct. While consistent with the virtual bargaining account, accounting for these results may be challenging for some accounts of pragmatic inference.

Collective patterns and stable misunderstandings in networks striving for consensus without a common value system

Collective phenomena in systems of interacting agents have helped us understand diverse social, ecological and biological observations. The corresponding explanations are challenged by incorrect information processing. In particular, the models typically assume a shared understanding of signals or a common truth or value system, i.e., an agreement of whether the measurement or perception of information is ‘right’ or ‘wrong’. It is an open question whether a collective consensus can emerge without these conditions. Here we introduce a model of interacting agents that strive for consensus, however, each with only a subjective perception of the world. Our communication model does not presuppose a definition of right or wrong and the actors can hence not distinguish between correct and incorrect observations. Depending on a single parameter that governs how responsive the agents are to changing their world-view we observe a transition between an unordered phase of individuals that are not able to communicate with each other and a phase of an emerging shared signalling framework. We find that there are two types of convention-aligned clusters: one, where all social actors in the cluster have the same set of conventions, and one, where neighbouring actors have different but compatible conventions (‘stable misunderstandings’).

Simplicity and informativeness in semantic category systems

Recent research has shown that semantic category systems, such as color and kinship terms, find an optimal balance between simplicity and informativeness. We argue that this situation arises through pressure for simplicity from learning and pressure for informativeness from communicative interaction, two distinct pressures that often (but not always) pull in opposite directions. Another account argues that learning might also act as a pressure for informativeness, that learners might be biased toward inferring informative systems. This results in two competing hypotheses about the human inductive bias. We formalize these competing hypotheses in a Bayesian iterated learning model in order to simulate what kinds of languages are expected to emerge under each. We then test this model experimentally to investigate whether learners' biases, isolated from any communicative task, are better characterized as favoring simplicity or informativeness. We find strong evidence to support the simplicity account. Furthermore, we show how the application of a simplicity principle in learning can give the impression of a bias for informativeness, even when no such bias is present. Our findings suggest that semantic categories are learned through domain-general principles, negating the need to posit a domain-specific mechanism.

From Context to Code: Information Transfer Constrains the Emergence of Graphic Codes

Humans commit information to graphic symbols for three basic reasons: as a memory aid, as a tool for thinking, and as a means of communication. Yet, despite the benefits of transmitting information graphically, we still know very little about the biases and constraints acting on the emergence of stable, powerful, and accurate graphic codes (such as writing). Using a reference game, where participants play as Messengers and Recipients, we experimentally manipulate the function of the task (communicative or non-communicative) and investigate whether this shapes the emergence of stable, powerful, and accurate codes for both synchronous and asynchronous modes of information transfer. Only in the Dialogue condition, where Messenger and Recipient are two different persons communicating within the same time frame (i.e., synchronously), do we consistently observe the emergence of stable, powerful, and accurate graphic codes. Such codes are unnecessary for participants in Recall, where Messenger and Recipient are the same person transferring information within the same time frame, and they fail to emerge in Correspondence, where Messenger and Recipient are two different persons communicating across time frames (i.e., asynchronously). Lastly, in the Mnemonic condition, where Messenger and Recipient are the same person at different points in time, participants achieve high accuracy but with codes that are suboptimal in terms of power and stability. Our results suggest that the rarity and late arrival of stable, powerful, and accurate graphic codes in human history largely stems from strong constraints on information transfer. In particular, we suggest that these constraints limit a code's ability to reach an adequate tradeoff between information that needs to be explicitly encoded and information that needs to be inferred from context.

The Emergence of Social Norms and Conventions

The utility of our actions frequently depends upon the beliefs and behavior of other agents. Thankfully, through experience, we learn norms and conventions that provide stable expectations for navigating our social world. Here, we review several distinct influences on their content and distribution. At the level of individuals locally interacting in dyads, success depends on rapidly adapting pre-existing norms to the local context. Hence, norms are shaped by complex cognitive processes involved in learning and social reasoning. At the population level, norms are influenced by intergenerational transmission and the structure of the social network. As human social connectivity continues to increase, understanding and predicting how these levels and time scales interact to produce new norms will be crucial for improving communities.

Emergence of linguistic conventions in multi-agent reinforcement learning

Recently, emergence of signaling conventions, among which language is a prime example, draws a considerable interdisciplinary interest ranging from game theory, to robotics to evolutionary linguistics. Such a wide spectrum of research is based on much different assumptions and methodologies, but complexity of the problem precludes formulation of a unifying and commonly accepted explanation. We examine formation of signaling conventions in a framework of a multi-agent reinforcement learning model. When the network of interactions between agents is a complete graph or a sufficiently dense random graph, a global consensus is typically reached with the emerging language being a nearly unique object-word mapping or containing some synonyms and homonyms. On finite-dimensional lattices, the model gets trapped in disordered configurations with a local consensus only. Such a trapping can be avoided by introducing a population renewal, which in the presence of superlinear reinforcement restores an ordinary surface-tension driven coarsening and considerably enhances formation of efficient signaling.

Coevolution of Lexical Meaning and Pragmatic Use

According to standard linguistic theory, the meaning of an utterance is the product of conventional semantic meaning and general pragmatic rules on language use. We investigate how such a division of labor between semantics and pragmatics could evolve under general processes of selection and learning. We present a game‐theoretic model of the competition between types of language users, each endowed with certain lexical representations and a particular pragmatic disposition to act on them. Our model traces two evolutionary forces and their interaction: (i) pressure toward communicative efficiency and (ii) transmission perturbations during the acquisition of linguistic knowledge. We illustrate the model based on a case study on scalar implicatures, which suggests that the relationship between underspecified semantics and pragmatic inference is one of coevolution.