We used a non-linguistic experimental paradigm to explore the instantaneous creation of new communicative con-ventions. Participants played a computer game, in which they sent and interpreted minimal signals to obtain shared rewardswithin a virtual scene. Trials manipulated the space of possible signals that could be sent, and the meanings to be expressed (lo-cations and quantities of rewards); as such, optimal success in the task required participants to jointly construct signal-meaningmappings that functioned as part of a system, rather than in isolation.We observed different signalling strategies among participants, but with some individuals using ‘system-mapping’ conven-tions that globally reorganized in light of changing task constraints. Such behaviour reflects the principle of pre-emption inpragmatics, where the inferred meaning of an utterance depends on its relationship among a set of alternatives. Our initialfindings provide a basis for future research, investigating contexts that are conducive to this phenomenon.

We used a novel experimental paradigm to investigate cognitive principles underpinning human communication. Through a computer game simulating different virtual scenes, pairs of participants sent and interpreted non-linguistic, minimal signals to achieve common goals. Participants’ signalling and interpretative actions demonstrated both flexibility and sensitivity to variations in the context of the shared visual scene: the same signal in one context could ‘flip’ its meaning in a new context. Such ‘invertible’ signals in the lab have their counterparts in patterns of real-world natural language use—from the phenomenon of enantiosemy (words/phrases that contain their ‘opposite’ meaning) to the pragmatics of satire and irony. But the emergence of such signals in our experiment challenges both correlational associative and recursive-mentalizing (‘mind-reading’) accounts of human communication and language. Instead, we point to a pragmatics-central perspective in which what is vital is our capacity for joint inference and coordination.

Human estimates of the probabilities of combinations of events show well-established violations of probabilitytheory, most notably the conjunction and disjunction fallacies. These violations have led researchers to conclude that therules of probability are too complex for most people to use, and that cognitively-easier approximations such as averagingare used instead. Unlike previous work that has assumed that individuals use only a single combination rule, we collectedrepeated estimates of conjunctions and disjunctions and investigated whether individuals consistently used a single rule orused a repertoire of rules in a trial-by-trial Bayesian analysis. When not incentivized, most participants were best describedas randomly selecting a combination rule on each trial, and the correct rule was the most often used. Despite this, whenincentivized to use their single-best strategy participants were more likely to use the incorrect averaging rule. People do notseem to understand their own strategies well.

Do people use a single strategy or sample from multiple strategies when estimating the conjunction and disjunction of two independent events? Here we address this tension directly by comparing individual level Bayesian simulations of multi and single strategy models using data from a frequency estimation experiment. Participants were shown two statements describing attributes and asked to estimate how many people had either one attribute, conjunction, or disjunction of attributes. In our Bayesian simulations we compare models in which participants either adopt a single strategy or sample from a set of strategies when forming estimates of both conjunctions and disjunctions. We compared every permutation of models in which a participant is responding based on a single component, a weighted average of the two events, probability theory, or combination of strategies. Our findings show that people sample from multiple strategies and are sampling from the normative strategy some of the time.

When asked to combine two pieces of evidence, one diagnostic and one non-diagnostic, people show a dilution effect: the addition of non-diagnostic evidence dilutes the overall strength of the evidence. This non-normative effect has been found in a variety of tasks and has been taken as evidence that people inappropriately combine information. We investigated the dilution effect using simple perceptual stimuli but unlike in previous work we asked participants to judge likelihoods ratios, allowing us to assess not just ordinal relationship between judgments but also whether each individual judgment was accurate. We found the dilution effect, but surprisingly it was not due to inaccurate combination of diagnostic and non-diagnostic information. People were accurate at judging diagnostic evidence combined with non-diagnostic evidence, but overestimated the strength of diagnostic evidence alone. We explain this within-participants dilution effect as the result of inference about missing features rather than incorrect combination of information.

The development and application of concepts is a critical component of cognition. Although concepts can be formed on the basis of simple perceptual or semantic features, conceptual representations can also capitalize on similarities across feature relationships. By representing these types of higher-order relationships, concepts can simplify the learning problem and facilitate decisions. Despite this, little is known about the neural mechanisms that support the construction and deployment of these kinds of higher-order concepts during learning. To address this question, we combined a carefully designed associative learning task with computational model-based functional magnetic resonance imaging (fMRI). Participants were scanned as they learned and made decisions about sixteen pairs of cues and associated outcomes. Associations were structured such that individual cues shared feature relationships, operationalized as shared patterns of cue pair-outcome associations. In order to capture the large number of possible conceptual representational structures that participants might employ and to evaluate how conceptual representations are used during learning, we leveraged a well-specified Bayesian computational model of category learning [1]. Behavioral and model-based results revealed that participants who displayed a tendency to link experiences in memory benefitted from faster learning rates, suggesting that the use of the conceptual structure in the task facilitated decisions about cue pair-outcome associations. Model-based fMRI analyses revealed that trial-by-trial integration of cue information into higher-order conceptual representations was supported by an anterior temporal (AT) network of regions previously implicated in representing complex conjunctions of features and meaning-based information.