Shared functional specialization in transformer-based language models and the human brain.

When processing language, the brain is thought to deploy specialized computations to construct meaning from complex linguistic structures. Recently, artificial neural networks based on the Transformer architecture have revolutionized the field of natural language processing. Transformers integrate contextual information across words via structured circuit computations. Prior work has focused on the internal representations ("embeddings") generated by these circuits. In this paper, we instead analyze the circuit computations directly: we deconstruct these computations into the functionally-specialized "transformations" that integrate contextual information across words. Using functional MRI data acquired while participants listened to naturalistic stories, we first verify that the transformations account for considerable variance in brain activity across the cortical language network. We then demonstrate that the emergent computations performed by individual, functionally-specialized "attention heads" differentially predict brain activity in specific cortical regions. These heads fall along gradients corresponding to different layers and context lengths in a low-dimensional cortical space.

A Hierarchical Bayesian Model of Adaptive Teaching.

How do teachers learn about what learners already know? How do learners aid teachers by providing them with information about their background knowledge and what they find confusing? We formalize this collaborative reasoning process using a hierarchical Bayesian model of pedagogy. We then evaluate this model in two online behavioral experiments (N = 312 adults). In Experiment 1, we show that teachers select examples that account for learners' background knowledge, and adjust their examples based on learners' feedback. In Experiment 2, we show that learners strategically provide more feedback when teachers' examples deviate from their background knowledge. These findings provide a foundation for extending computational accounts of pedagogy to richer interactive settings.

Interaction structure constrains the emergence of conventions in group communication.

Real-world communication frequently requires language producers to address more than one comprehender at once, yet most psycholinguistic research focuses on one-on-one communication. As the audience size grows, interlocutors face new challenges that do not arise in dyads. They must consider multiple perspectives and weigh multiple sources of feedback to build shared understanding. Here, we ask which properties of the group's interaction structure facilitate successful communication. We used a repeated reference game paradigm in which directors instructed between one and five matchers to choose specific targets out of a set of abstract figures. Across 313 games (N = 1,319 participants), we manipulated several key constraints on the group's interaction, including the amount of feedback that matchers could give to directors and the availability of peer interaction between matchers. Across groups of different sizes and interaction constraints, describers produced increasingly efficient utterances and matchers made increasingly accurate selections. Critically, however, we found that smaller groups and groups with less-constrained interaction structures ("thick channels") showed stronger convergence to group-specific conventions than large groups with constrained interaction structures ("thin channels"), which struggled with convention formation. Overall, these results shed light on the core structural factors that enable communication to thrive in larger groups.

Single-neuron analysis of aging-associated changes in learning reveals impairments in transcriptional plasticity.

The molecular mechanisms underlying age-related declines in learning and long-term memory are still not fully understood. To address this gap, our study focused on investigating the transcriptional landscape of a singularly identified motor neuron L7 in Aplysia, which is pivotal in a specific type of nonassociative learning known as sensitization of the siphon-withdraw reflex. Employing total RNAseq analysis on a single isolated L7 motor neuron after short-term or long-term sensitization (LTS) training of Aplysia at 8, 10, and 12 months (representing mature, late mature, and senescent stages), we uncovered aberrant changes in transcriptional plasticity during the aging process. Our findings specifically highlight changes in the expression of messenger RNAs (mRNAs) that encode transcription factors, translation regulators, RNA methylation participants, and contributors to cytoskeletal rearrangements during learning and long noncoding RNAs (lncRNAs). Furthermore, our comparative gene expression analysis identified distinct transcriptional alterations in two other neurons, namely the motor neuron L11 and the giant cholinergic neuron R2, whose roles in LTS are not yet fully elucidated. Taken together, our analyses underscore cell type-specific impairments in the expression of key components related to learning and memory within the transcriptome as organisms age, shedding light on the complex molecular mechanisms driving cognitive decline during aging.

The Emergence of Specialized Roles Within Groups.

Humans routinely form groups to achieve goals that no individual can accomplish alone. Group coordination often brings to mind synchrony and alignment, where all individuals do the same thing (e.g., driving on the right side of the road, marching in lockstep, or playing musical instruments on a regular beat). Yet, effective coordination also typically involves differentiation, where specialized roles emerge for different members (e.g., prep stations in a kitchen or positions on an athletic team). Role specialization poses a challenge for computational models of group coordination, which have largely focused on achieving synchrony. Here, we present the CARMI framework, which characterizes role specialization processes in terms of five core features that we hope will help guide future model development: Communication, Adaptation to feedback, Repulsion, Multi-level planning, and Intention modeling. Although there are many paths to role formation, we suggest that roles emerge when each agent in a group dynamically allocates their behavior toward a shared goal to complement what they expect others to do. In other words, coordination concerns beliefs (who will do what) rather than simple actions. We describe three related experimental paradigms-"Group Binary Search," "Battles of the Exes," and "Find the Unicorn"-that we have used to study differentiation processes in the lab, each emphasizing different aspects of the CARMI framework.

Reconciling truthfulness and relevance as epistemic and decision-theoretic utility.

People use language to influence others' beliefs and actions. Yet models of communication have diverged along these lines, formalizing the speaker's objective in terms of either the listener's beliefs or actions. We argue that this divergence lies at the root of a longstanding controversy over the Gricean maxims of truthfulness and relevance. We first bridge the divide by introducing a speaker model which considers both the listener's beliefs (epistemic utility) and their actions (decision-theoretic utility). We show that formalizing truthfulness as an epistemic utility and relevance as a decision-theoretic utility reconciles the tension between them, readily explaining puzzles such as context-dependent standards of truthfulness. We then test a set of novel predictions generated by our model. We introduce a new signaling game which decouples utterances' truthfulness and relevance, then use it to conduct a pair of experiments. Our first experiment demonstrates that participants jointly maximize epistemic and decision-theoretic utility, rather than either alone. Our second experiment shows that when the two conflict, participants make a graded tradeoff rather than prioritizing one over the other. These results demonstrate that human communication cannot be reduced to influencing beliefs or actions alone. Taken together, our work provides a new foundation for grounding rational communication not only in what we believe, but in what those beliefs lead us to do. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Flexible social inference facilitates targeted social learning when rewards are not observable.

Groups coordinate more effectively when individuals are able to learn from others' successes. But acquiring such knowledge is not always easy, especially in real-world environments where success is hidden from public view. We suggest that social inference capacities may help bridge this gap, allowing individuals to update their beliefs about others' underlying knowledge and success from observable trajectories of behaviour. We compared our social inference model against simpler heuristics in three studies of human behaviour in a collective-sensing task. Experiment 1 demonstrated that average performance improved as a function of group size at a rate greater than predicted by heuristic models. Experiment 2 introduced artificial agents to evaluate how individuals selectively rely on social information. Experiment 3 generalized these findings to a more complex reward landscape. Taken together, our findings provide insight into the relationship between individual social cognition and the flexibility of collective behaviour.

Distribution, cellular localization, and colocalization of several peptide neurotransmitters in the central nervous system of Aplysia.

Neuropeptides are widely used as neurotransmitters in vertebrates and invertebrates. In vertebrates, a detailed understanding of their functions as transmitters has been hampered by the complexity of the nervous system. The marine mollusk Aplysia, with a simpler nervous system and many large, identified neurons, presents several advantages for addressing this question and has been used to examine the roles of tens of peptides in behavior. To screen for other peptides that might also play roles in behavior, we observed immunoreactivity in individual neurons in the central nervous system of adult Aplysia with antisera raised against the Aplysia peptide FMRFamide and two mammalian peptides that are also found in Aplysia, cholecystokinin (CCK) and neuropeptide Y (NPY), as well as serotonin (5HT). In addition, we observed staining of individual neurons with antisera raised against mammalian somatostatin (SOM) and peptide histidine isoleucine (PHI). However, genomic analysis has shown that these two peptides are not expressed in the Aplysia nervous system, and we have therefore labeled the unknown peptides stained by these two antibodies as XSOM and XPHI There was an area at the anterior end of the cerebral ganglion that had staining by antisera raised against many different transmitters, suggesting that this may be a modulatory region of the nervous system. There was also staining for XSOM and, in some cases, FMRFamide in the bag cell cluster of the abdominal ganglion. In addition, these and other studies have revealed a fairly high degree of colocalization of different neuropeptides in individual neurons, suggesting that the peptides do not just act independently but can also interact in different combinations to produce complex functions. The simple nervous system of Aplysia is advantageous for further testing these ideas.

Visual resemblance and interaction history jointly constrain pictorial meaning.

How do drawings-ranging from detailed illustrations to schematic diagrams-reliably convey meaning? Do viewers understand drawings based on how strongly they resemble an entity (i.e., as images) or based on socially mediated conventions (i.e., as symbols)? Here we evaluate a cognitive account of pictorial meaning in which visual and social information jointly support visual communication. Pairs of participants used drawings to repeatedly communicate the identity of a target object among multiple distractor objects. We manipulated social cues across three experiments and a full replication, finding that participants developed object-specific and interaction-specific strategies for communicating more efficiently over time, beyond what task practice or a resemblance-based account alone could explain. Leveraging model-based image analyses and crowdsourced annotations, we further determined that drawings did not drift toward "arbitrariness," as predicted by a pure convention-based account, but preserved visually diagnostic features. Taken together, these findings advance psychological theories of how successful graphical conventions emerge.

Show or tell? Exploring when (and why) teaching with language outperforms demonstration.

People use a wide range of communicative acts across different modalities, from concrete demonstrations to abstract language. While these modalities are typically studied independently, we take a comparative approach and ask when and why one modality might outperform another. We present a series of real-time, multi-player experiments asking participants to teach concepts using either demonstrations or language. Our first experiment (N=416) asks when language might outperform demonstration. We manipulate the complexity of the concept being taught and find that language communicates complex concepts more effectively than demonstration. We then ask why language succeeds in this setting. We hypothesized that language allowed teachers to reference abstract object features (e.g., shapes and colors), while demonstration teachers could only provide concrete examples (specific positive or negative objects). To test this hypothesis, our second experiment (N=568) ablated object features from the teacher's interface. This manipulation severely impaired linguistic (but not demonstrative) teaching. Our findings suggest that language communicates complex concepts by directly transmitting abstract rules. In contrast, demonstrations transmit examples, requiring the learner to infer the rules.

From partners to populations: A hierarchical Bayesian account of coordination and convention.

Languages are powerful solutions to coordination problems: They provide stable, shared expectations about how the words we say correspond to the beliefs and intentions in our heads. Yet, language use in a variable and nonstationary social environment requires linguistic representations to be flexible: Old words acquire new ad hoc or partner-specific meanings on the fly. In this article, we introduce continual hierarchical adaptation through inference (CHAI), a hierarchical Bayesian theory of coordination and convention formation that aims to reconcile the long-standing tension between these two basic observations. We argue that the central computational problem of communication is not simply transmission, as in classical formulations, but continual learning and adaptation over multiple timescales. Partner-specific common ground quickly emerges from social inferences within dyadic interactions, while community-wide social conventions are stable priors that have been abstracted away from interactions with multiple partners. We present new empirical data alongside simulations showing how our model provides a computational foundation for several phenomena that have posed a challenge for previous accounts: (a) the convergence to more efficient referring expressions across repeated interaction with the same partner, (b) the gradual transfer of partner-specific common ground to strangers, and (c) the influence of communicative context on which conventions eventually form. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

A Pragmatic Account of the Weak Evidence Effect.

Language is not only used to transmit neutral information; we often seek to persuade by arguing in favor of a particular view. Persuasion raises a number of challenges for classical accounts of belief updating, as information cannot be taken at face value. How should listeners account for a speaker's "hidden agenda" when incorporating new information? Here, we extend recent probabilistic models of recursive social reasoning to allow for persuasive goals and show that our model provides a pragmatic account for why weakly favorable arguments may backfire, a phenomenon known as the weak evidence effect. Critically, this model predicts a systematic relationship between belief updates and expectations about the information source: weak evidence should only backfire when speakers are expected to act under persuasive goals and prefer the strongest evidence. We introduce a simple experimental paradigm called the Stick Contest to measure the extent to which the weak evidence effect depends on speaker expectations, and show that a pragmatic listener model accounts for the empirical data better than alternative models. Our findings suggest further avenues for rational models of social reasoning to illuminate classical decision-making phenomena.

Semantic relatedness retroactively boosts memory and promotes memory interdependence across episodes.

Two fundamental issues in memory research concern when later experiences strengthen or weaken initial memories and when the two memories become linked or remain independent. A promising candidate for explaining these issues is semantic relatedness. Here, across five paired-associate learning experiments (N=1000), we systematically varied the semantic relatedness between initial and later cues, initial and later targets, or both. We found that learning retroactively benefited long-term memory performance for semantically related words (vs. unshown control words), and these benefits increased as a function of relatedness. Critically, memory dependence between initial and later pairs also increased with relatedness, suggesting that pre-existing semantic relationships promote interdependence for memories formed across episodes. We also found that modest retroactive benefits, but not interdependencies, emerged when subjects learned via studying rather than practice testing. These findings demonstrate that semantic relatedness during new learning retroactively strengthens old associations while scaffolding new ones into well-fortified memory traces.

Shades of confusion: Lexical uncertainty modulates ad hoc coordination in an interactive communication task.

There is substantial variability in the expectations that communication partners bring into interactions, creating the potential for misunderstandings. To directly probe these gaps and our ability to overcome them, we propose a communication task based on color-concept associations. In Experiment 1, we establish several key properties of the mental representations of these expectations, or lexical priors, based on recent probabilistic theories. Associations are more variable for abstract concepts, variability is represented as uncertainty within each individual, and uncertainty enables accurate predictions about whether others are likely to share the same association. In Experiment 2, we then examine the downstream consequences of these representations for communication. Accuracy is initially low when communicating about concepts with more variable associations, but rapidly increases as participants form ad hoc conventions. Together, our findings suggest that people cope with variability by maintaining well-calibrated uncertainty about their partner and appropriately adaptable representations of their own.

Possible novel features of synaptic regulation during long-term facilitation in Aplysia.

Most studies of molecular mechanisms of synaptic plasticity have focused on the sequence of changes either at individual synapses or in the cell nucleus. However, studies of long-term facilitation at Aplysia sensory neuron-motor neuron synapses in isolated cell culture suggest two additional features of facilitation. First, that there is also regulation of the number of synaptic contacts between two neurons, which may occur at the level of cell pair-specific branch points in the neuronal arbor. Branch points contain many molecules that are involved in protein synthesis-dependent long-term facilitation including neurotrophins and the RNA binding protein CPEB. Second, the regulation involves homeostatic feedback and tends to keep the total number of contacts between two neurons at a fairly constant level both at rest and following facilitation. That raises the question of how facilitation and homeostasis can coexist. A possible answer is suggested by the findings that they both involve spontaneous transmission and postsynaptic Ca2+, which can have bidirectional effects similar to LTP and LTD in hippocampus. In addition, long-term facilitation can involve a change in the set point of homeostasis, which could be encoded by plasticity molecules such as CPEB and/or PKM. A computational model based on these ideas can qualitatively simulate the basic features of both facilitation and homeostasis of the number of contacts.

The Division of Labor in Communication: Speakers Help Listeners Account for Asymmetries in Visual Perspective.

Recent debates over adults' theory of mind use have been fueled by surprising failures of perspective-taking in communication, suggesting that perspective-taking may be relatively effortful. Yet adults routinely engage in effortful processes when needed. How, then, should speakers and listeners allocate their resources to achieve successful communication? We begin with the observation that the shared goal of communication induces a natural division of labor: The resources one agent chooses to allocate toward perspective-taking should depend on their expectations about the other's allocation. We formalize this idea in a resource-rational model augmenting recent probabilistic weighting accounts with a mechanism for (costly) control over the degree of perspective-taking. In a series of simulations, we first derive an intermediate degree of perspective weighting as an optimal trade-off between expected costs and benefits of perspective-taking. We then present two behavioral experiments testing novel predictions of our model. In Experiment 1, we manipulated the presence or absence of occlusions in a director-matcher task. We found that speakers spontaneously modulated the informativeness of their descriptions to account for "known unknowns" in their partner's private view, reflecting a higher degree of speaker perspective-taking than previously acknowledged. In Experiment 2, we then compared the scripted utterances used by confederates in prior work with those produced in interactions with unscripted directors. We found that confederates were systematically less informative than listeners would initially expect given the presence of occlusions, but listeners used violations to adaptively make fewer errors over time. Taken together, our work suggests that people are not simply "mindblind"; they use contextually appropriate expectations to navigate the division of labor with their partner. We discuss how a resource-rational framework may provide a more deeply explanatory foundation for understanding flexible perspective-taking under processing constraints.

Characterizing the Dynamics of Learning in Repeated Reference Games.

The language we use over the course of conversation changes as we establish common ground and learn what our partner finds meaningful. Here we draw upon recent advances in natural language processing to provide a finer-grained characterization of the dynamics of this learning process. We release an open corpus (>15,000 utterances) of extended dyadic interactions in a classic repeated reference game task where pairs of participants had to coordinate on how to refer to initially difficult-to-describe tangram stimuli. We find that different pairs discover a wide variety of idiosyncratic but efficient and stable solutions to the problem of reference. Furthermore, these conventions are shaped by the communicative context: words that are more discriminative in the initial context (i.e., that are used for one target more than others) are more likely to persist through the final repetition. Finally, we find systematic structure in how a speaker's referring expressions become more efficient over time: Syntactic units drop out in clusters following positive feedback from the listener, eventually leaving short labels containing open-class parts of speech. These findings provide a higher resolution look at the quantitative dynamics of ad hoc convention formation and support further development of computational models of learning in communication.

When redundancy is useful: A Bayesian approach to "overinformative" referring expressions.

Referring is one of the most basic and prevalent uses of language. How do speakers choose from the wealth of referring expressions at their disposal? Rational theories of language use have come under attack for decades for not being able to account for the seemingly irrational overinformativeness ubiquitous in referring expressions. Here we present a novel production model of referring expressions within the Rational Speech Act framework that treats speakers as agents that rationally trade off cost and informativeness of utterances. Crucially, we relax the assumption that informativeness is computed with respect to a deterministic Boolean semantics, in favor of a nondeterministic continuous semantics. This innovation allows us to capture a large number of seemingly disparate phenomena within one unified framework: the basic asymmetry in speakers' propensity to overmodify with color rather than size; the increase in overmodification in complex scenes; the increase in overmodification with atypical features; and the increase in specificity in nominal reference as a function of typicality. These findings cast a new light on the production of referring expressions: rather than being wastefully overinformative, reference is usefully redundant. (PsycInfo Database Record (c) 2020 APA, all rights reserved).

Comparison of the ionic currents modulated during activity-dependent and normal presynaptic facilitation.

One of the major questions in psychology is whether associative and nonassociative learning are fundamentally different or whether they involve similar processes and mechanisms. We have addressed this question by comparing mechanisms of a nonassociative form of learning, sensitization, and an associative form of learning, classical conditioning of the siphon-withdrawal reflex of hermaphroditic Aplysia In an analog of differential conditioning, action potentials in one siphon sensory neuron (SN) were paired with shock to the pedal nerves, producing activity-dependent presynaptic facilitation, and action potentials in another SN were unpaired with the shock as a control. The difference between paired and unpaired training is a measure of associative plasticity. Before and after this training, we voltage clamped each SN and measured the outward current during depolarizing pulses. There was a significantly greater decrease in the net outward current in the paired SN than in the unpaired SN. We obtained similar results when we substituted the depolarizing voltage clamp pulse for action potentials during training. We then bathed the ganglion in serotonin as a measure of nonassociative plasticity. The current that was modulated differentially (paired-unpaired) had time and voltage dependencies similar to the current that was modulated by serotonin (I s). These results suggest that an associative form of plasticity, activity-dependent presynaptic facilitation underlying conditioning, involves enhanced modulation of the same ionic current as a nonassociative form, normal presynaptic facilitation underlying sensitization.

The contributions and mechanisms of changes in excitability during simple forms of learning in Aplysia.

Learning and memory have long been thought to involve changes in synaptic connections between neurons. However, in many cases learning-related plasticity also involves changes in the excitability of neurons. These findings have raised questions about the relative importance of these two types of mechanisms to behavioral learning, and also about the extent to which they involve shared or unique molecular mechanisms. We have taken a reductionist approach to these questions by addressing them in a simple model organism, Aplysia californica. Studies of a semi-intact Aplysia siphon withdrawal preparation suggest that classical conditioning involves an increase in the evoked firing of sensory neurons (SNs) as well as facilitation of the monosynaptic PSP to motor neurons (MNs). Furthermore, these two mechanisms may act cooperatively at the cellular level: increased SN firing produces more PSPs, each of which is facilitated, leading to a multiplicative increase in depolarization of the MN and siphon withdrawal. The changes in SN firing and the monosynaptic PSP also share several mechanisms at the molecular level, suggesting that they may both be due in part to a decrease in K+ current that causes an increase in SN excitability as well as an increase in SN spike width and thus increased transmitter release. However, changes in the monosynaptic PSP also involve additional mechanisms that are not shared and may affect different aspects of synaptic transmission as well. Studies of operant conditioning of feeding suggest that it involves similar mechanisms as classical conditioning of siphon withdrawal. In particular, for both types of associative learning adenylyl cyclase appears to serve as a molecular coincidence detector that leads to increased activation of PKA and changes in excitability of key neurons in the neural circuit. Furthermore, in both cases those changes in excitability make an important contribution to the behavioral learning.