An algorithmic account for how humans efficiently learn, transfer, and compose hierarchically structured decision policies.

Learning structures that effectively abstract decision policies is key to the flexibility of human intelligence. Previous work has shown that humans use hierarchically structured policies to efficiently navigate complex and dynamic environments. However, the computational processes that support the learning and construction of such policies remain insufficiently understood. To address this question, we tested 1026 human participants, who made over 1 million choices combined, in a decision-making task where they could learn, transfer, and recompose multiple sets of hierarchical policies. We propose a novel algorithmic account for the learning processes underlying observed human behavior. We show that humans rely on compressed policies over states in early learning, which gradually unfold into hierarchical representations via meta-learning and Bayesian inference. Our modeling evidence suggests that these hierarchical policies are structured in a temporally backward, rather than forward, fashion. Taken together, these algorithmic architectures characterize how the interplay between reinforcement learning, policy compression, meta-learning, and working memory supports structured decision-making and compositionality in a resource-rational way.

Group Coordination Catalyzes Individual and Cultural Intelligence.

A large program of research has aimed to ground large-scale cultural phenomena in processes taking place within individual minds. For example, investigating whether individual agents equipped with the right social learning strategies can enable cumulative cultural evolution given long enough time horizons. However, this approach often omits the critical group-level processes that mediate between individual agents and multi-generational societies. Here, we argue that interacting groups are a necessary and explanatory level of analysis, linking individual and collective intelligence through two characteristic feedback loops. In the first loop, more sophisticated individual-level social learning mechanisms based on Theory of Mind facilitate group-level complementarity, allowing distributed knowledge to be compositionally recombined in groups; these group-level innovations, in turn, ease the cognitive load on individuals. In the second loop, societal-level processes of cumulative culture provide groups with new cognitive technologies, including shared language and conceptual abstractions, which set in motion new group-level processes to further coordinate, recombine, and innovate. Taken together, these cycles establish group-level interaction as a dual engine of intelligence, catalyzing both individual cognition and cumulative culture.

A Phone in a Basket Looks Like a Knife in a Cup: Role-Filler Independence in Visual Processing.

When a piece of fruit is in a bowl, and the bowl is on a table, we appreciate not only the individual objects and their features, but also the relations containment and support, which abstract away from the particular objects involved. Independent representation of roles (e.g., containers vs. supporters) and "fillers" of those roles (e.g., bowls vs. cups, tables vs. chairs) is a core principle of language and higher-level reasoning. But does such role-filler independence also arise in automatic visual processing? Here, we show that it does, by exploring a surprising error that such independence can produce. In four experiments, participants saw a stream of images containing different objects arranged in force-dynamic relations-e.g., a phone contained in a basket, a marker resting on a garbage can, or a knife sitting in a cup. Participants had to respond to a single target image (e.g., a phone in a basket) within a stream of distractors presented under time constraints. Surprisingly, even though participants completed this task quickly and accurately, they false-alarmed more often to images matching the target's relational category than to those that did not-even when those images involved completely different objects. In other words, participants searching for a phone in a basket were more likely to mistakenly respond to a knife in a cup than to a marker on a garbage can. Follow-up experiments ruled out strategic responses and also controlled for various confounding image features. We suggest that visual processing represents relations abstractly, in ways that separate roles from fillers.

Early-emerging combinatorial thought: Human infants flexibly combine kind and quantity concepts.

Combinatorial thought, or the ability to combine a finite set of concepts into a myriad of complex ideas and knowledge structures, is the key to the productivity of the human mind and underlies communication, science, technology, and art. Despite the importance of combinatorial thought for human cognition and culture, its developmental origins remain unknown. To address this, we tested whether 12-mo-old infants (N = 60), who cannot yet speak and only understand a handful of words, can combine quantity and kind concepts activated by verbal input. We proceeded in two steps: first, we taught infants two novel labels denoting quantity (e.g., "mize" for 1 item; "padu" for 2 items, Experiment 1). Then, we assessed whether they could combine quantity and kind concepts upon hearing complex expressions comprising their labels (e.g., "padu duck", Experiments 2-3). At test, infants viewed four different sets of objects (e.g., 1 duck, 2 ducks, 1 ball, 2 balls) while being presented with the target phrase (e.g., "padu duck") naming one of them (e.g., 2 ducks). They successfully retrieved and combined on-line the labeled concepts, as evidenced by increased looking to the named sets but not to distractor sets. Our results suggest that combinatorial processes for building complex representations are available by the end of the first year of life. The infant mind seems geared to integrate concepts in novel productive ways. This ability may be a precondition for deciphering the ambient language(s) and building abstract models of experience that enable fast and flexible learning.

Compositionality in perception: A framework.

Perception involves the processing of content or information about the world. In what form is this content represented? I argue that perception is widely compositional. The perceptual system represents many stimulus features (including shape, orientation, and motion) in terms of combinations of other features (such as shape parts, slant and tilt, common and residual motion vectors). But compositionality can take a variety of forms. The ways in which perceptual representations compose are markedly different from the ways in which sentences or thoughts are thought to be composed. I suggest that the thesis that perception is compositional is not itself a concrete hypothesis with specific predictions; rather it affords a productive framework for developing and evaluating specific empirical hypotheses about the form and content of perceptual representations. The question is not just whether perception is compositional, but how. Answering this latter question can provide fundamental insights into perception. This article is categorized under: Philosophy > Representation Philosophy > Foundations of Cognitive Science Psychology > Perception and Psychophysics.

Minimal Compositionality versus Bird Implicatures: two theories of ABC-D sequences in Japanese tits.

It was argued in a series of experimental studies that Japanese tits (Parus minor) have an ABC call that has an alert function, a D call that has a recruitment function, and an ABC-D call that is compositionally derived from ABC and D, and has a mobbing function. A key conclusion was that ABC-D differs from the combination of separate utterances of ABC and of D (e.g. as played by distinct but close loudspeakers). While the logic of the argument is arguably sound, no explicit rule has been proposed to derive the meaning of ABC-D from that of its parts. We compare two analyses. One posits a limited instance of semantic compositionality ('Minimal Compositionality'); the other does without compositionality, but uses instead a more sophisticated pragmatics ('Bird Implicatures'). Minimal Compositionality takes the composition of ABC and D to deviate only minimally from what would be found with two independent utterances: ABC means that 'there is something that licenses an alert', D means that 'there is something that licenses recruitment', and ABC-D means that 'there is something that licenses both an alert and recruitment'. By contrast, ABC and D as independent utterances yield something weaker, namely: 'there is something that licenses an alert, and there is something that licenses recruitment', without any 'binding' across the two utterances. The second theory, Bird Implicatures, only requires that ABC-D should be more informative than ABC, and/or than D. It builds on the idea, proposed for several monkey species, that a less-informative call competes with a more informative one (the 'Informativity Principle'): when produced alone, ABC and D trigger an inference that ABC-D is false. We explain how both Minimal Compositionality and Bird Implicatures could have evolved, and we compare the predictions of the two theories. Finally, we extend the discussion to some chimpanzee and meerkat sequences that might raise related theoretical problems.

Tracking the neural codes for words and phrases during semantic composition, working-memory storage, and retrieval.

The ability to compose successive words into a meaningful phrase is a characteristic feature of human cognition, yet its neural mechanisms remain incompletely understood. Here, we analyze the cortical mechanisms of semantic composition using magnetoencephalography (MEG) while participants read one-word, two-word, and five-word noun phrases and compared them with a subsequent image. Decoding of MEG signals revealed three processing stages. During phrase comprehension, the representation of individual words was sustained for a variable duration depending on phrasal context. During the delay period, the word code was replaced by a working-memory code whose activation increased with semantic complexity. Finally, the speed and accuracy of retrieval depended on semantic complexity and was faster for surface than for deep semantic properties. In conclusion, we propose that the brain initially encodes phrases using factorized dimensions for successive words but later compresses them in working memory and requires a period of decompression to access them.

Inconsistent effects of components as evidence for non-compositionality in chimpanzee face-gesture combinations? A response to Oña et al (2019).

Using field observations from a sanctuary, Oña and colleagues (DOI: 10.7717/peerj.7623) investigated the semantics of face-gesture combinations in chimpanzees (Pan troglodytes). The response of the animals to these signals was encoded as a binary measure: positive interactions such as approaching or grooming were considered affiliative; ignoring or attacking was considered non-affiliative. The relevant signals are illustrated in Fig. 1 (https://doi.org/10.7717/peerj.7623/fig-1), together with the outcome in terms of average affiliativeness. The authors observe that there seems to be no systematicity in the way the faces modify the responses to the gestures, sometimes reducing affiliativeness, sometimes increasing it. A strong interpretation of this result would be that the meaning of a gesture-face combination cannot be derived from the meaning of the gesture and the meaning of the face, that is, the interpretation of chimpanzees' face-gesture combinations are non compositional in nature. We will revisit this conclusion: we will exhibit simple compositional systems which, after all, may be plausible. At the methodological level, we argue that it is critical to lay out the theoretical options explicitly for a complete comparison of their pros and cons.

Human arm redundancy: a new approach for the inverse kinematics problem.

The inverse kinematics (IK) problem addresses how both humans and robotic systems coordinate movement to resolve redundancy, as in the case of arm reaching where more degrees of freedom are available at the joint versus hand level. This work focuses on which coordinate frames best represent human movements, enabling the motor system to solve the IK problem in the presence of kinematic redundancies. We used a multi-dimensional sparse source separation method to derive sets of basis (or source) functions for both the task and joint spaces, with joint space represented by either absolute or anatomical joint angles. We assessed the similarities between joint and task sources in each of these joint representations, finding that the time-dependent profiles of the absolute reference frame's sources show greater similarity to corresponding sources in the task space. This result was found to be statistically significant. Our analysis suggests that the nervous system represents multi-joint arm movements using a limited number of basis functions, allowing for simple transformations between task and joint spaces. Additionally, joint space seems to be represented in an absolute reference frame to simplify the IK transformations, given redundancies. Further studies will assess this finding's generalizability and implications for neural control of movement.

Compositional diversity in visual concept learning.

Humans leverage compositionality to efficiently learn new concepts, understanding how familiar parts can combine together to form novel objects. In contrast, popular computer vision models struggle to make the same types of inferences, requiring more data and generalizing less flexibly than people do. Here, we study these distinctively human abilities across a range of different types of visual composition, examining how people classify and generate "alien figures" with rich relational structure. We also develop a Bayesian program induction model which searches for the best programs for generating the candidate visual figures, utilizing a large program space containing different compositional mechanisms and abstractions. In few shot classification tasks, we find that people and the program induction model can make a range of meaningful compositional generalizations, with the model providing a strong account of the experimental data as well as interpretable parameters that reveal human assumptions about the factors invariant to category membership (here, to rotation and changing part attachment). In few shot generation tasks, both people and the models are able to construct compelling novel examples, with people behaving in additional structured ways beyond the model capabilities, e.g. making choices that complete a set or reconfigure existing parts in new ways. To capture these additional behavioral patterns, we develop an alternative model based on neuro-symbolic program induction: this model also composes new concepts from existing parts yet, distinctively, it utilizes neural network modules to capture residual statistical structure. Together, our behavioral and computational findings show how people and models can produce a variety of compositional behavior when classifying and generating visual objects.

Sentiment annotations for 3827 simplified Chinese characters.

Sentiment analysis in Chinese natural language processing has been largely based on words annotated with sentiment categories or scores. Characters, however, are the basic orthographic, phonological, and in most cases, semantic units in the Chinese language. This study collected sentiment annotations for 3827 characters. The ratings demonstrated high levels of reliability, and were validated through a comparison with the ratings of some characters' word equivalents reported in a previous norming study. Relations with other lexico-semantic variables and character processing efficiency were investigated. Furthermore, analyses of the association between constituent character valence and word valence revealed semantic compositionality and sentiment fusion characteristic of larger Chinese linguistic units. These ratings for characters, expanding current Chinese sentiment lexicons, can be utilized for the purposes of more precise stimuli assessment in research on Chinese character processing and more efficient sentiment analysis equipped with annotations of single-character words.

Feline faces: Unraveling the social function of domestic cat facial signals.

Lately, there has been a growing interest in studying domestic cat facial signals, but most of this research has centered on signals produced during human-cat interactions or pain. The available research on intraspecific facial signaling with domesticated cats has largely focused on non-affiliative social interactions. However, the transition to intraspecific sociality through domestication could have resulted in a greater reliance on affiliative facial signals that aid with social bonding. Our study aimed to document the various facial signals that cats produce during affiliative and non-affiliative intraspecific interactions. Given the close relationship between the physical form and social function of mammalian facial signals, we predicted that affiliative and non-affiliative facial signals would have noticeable differences in their physical morphology. We observed the behavior of 53 adult domestic shorthair cats at CatCafé Lounge in Los Angeles, CA. Using Facial Action Coding Systems designed for cats, we compared the complexity and compositionality of facial signals produced in affiliative and non-affiliative contexts. To measure complexity and compositionality, we examined the number and types of facial muscle movements (AUs) observed in each signal. We found that compositionality, rather than complexity, was significantly associated with the social function of intraspecific facial signals. Our findings indicate that domestication likely had a significant impact on the development of intraspecific facial signaling repertoires in cats.

Overview of data preprocessing for machine learning applications in human microbiome research.

Although metagenomic sequencing is now the preferred technique to study microbiome-host interactions, analyzing and interpreting microbiome sequencing data presents challenges primarily attributed to the statistical specificities of the data (e.g., sparse, over-dispersed, compositional, inter-variable dependency). This mini review explores preprocessing and transformation methods applied in recent human microbiome studies to address microbiome data analysis challenges. Our results indicate a limited adoption of transformation methods targeting the statistical characteristics of microbiome sequencing data. Instead, there is a prevalent usage of relative and normalization-based transformations that do not specifically account for the specific attributes of microbiome data. The information on preprocessing and transformations applied to the data before analysis was incomplete or missing in many publications, leading to reproducibility concerns, comparability issues, and questionable results. We hope this mini review will provide researchers and newcomers to the field of human microbiome research with an up-to-date point of reference for various data transformation tools and assist them in choosing the most suitable transformation method based on their research questions, objectives, and data characteristics.

How Complex Verbs Acquire Their Idiosyncratic Meanings.

Complex verbs with the same preverb/prefix/particle that is both linguistically productive and analyzable can be compositional as well as non-compositional in meaning. For example, the English on has compositional spatial uses (put a hat on) but also a non-spatial "continuative" use, where its semantic contribution is consistent with multiple verbs (we played / worked / talked on despite the interruption). Comparable examples can be given with German preverbs or Russian prefixes, which are the main data analyzed in the present paper. The preverbs/prefixes/particles that encode non-compositional, construction-specific senses have been extensively studied; however, it is still far from clear how their semantic idiosyncrasies arise. Even when one can identify the contribution of the base, it is counterintuitive to assign the remaining sememes to the preverb/prefix/particle part. Therefore, on one hand, there seems to be an element without meaning, and on the other, there is a word sense that apparently comes from nowhere. In this article, I suggest analyzing compositional and non-compositional complex verbs as instantiations of two different types of constructions: one with an open slot for the preverb/prefix/particle and a fixed base verb and another with a fixed preverb/prefix/particle and an open slot for the base verb. Both experimental and corpus evidence supporting this decision is provided for Russian data. I argue that each construction implies its own meaning-processing model and that the actual choice between the two can be predicted by taking into account the discrepancy in probabilities of transition from preverb/prefix/particle to base and from base to preverb/prefix/particle.

The neural ingredients for a language of thought are available.

The classical notion of a 'language of thought' (LoT), advanced prominently by the philosopher Jerry Fodor, is an influential position in cognitive science whereby the mental representations underpinning thought are considered to be compositional and productive, enabling the construction of new complex thoughts from more primitive symbolic concepts. LoT theory has been challenged because a neural implementation has been deemed implausible. We disagree. Examples of critical computational ingredients needed for a neural implementation of a LoT have in fact been demonstrated, in particular in the hippocampal spatial navigation system of rodents. Here, we show that cell types found in spatial navigation (border cells, object cells, head-direction cells, etc.) provide key types of representation and computation required for the LoT, underscoring its neurobiological viability.

Visual Form and Event Semantics Predict Transitivity in Silent Gestures: Evidence for Compositionality.

Silent gesture is not considered to be linguistic, on par with spoken and sign languages. It is claimed that silent gestures, unlike language, represent events holistically, without compositional structure. However, recent research has demonstrated that gesturers use consistent strategies when representing objects and events, and that there are behavioral and clinically relevant limits on what form a gesture may take to effect a particular meaning. This systematicity challenges a holistic interpretation of silent gesture, which predicts that there should be no stable form-meaning correspondence across event representations. Here, we demonstrate to the contrary that untrained gesturers systematically manipulate the form of their gestures when representing events with and without a theme (e.g., Someone popped the balloon vs. Someone walked), that is, transitive and intransitive events. We elicited silent gestures and annotated them for manual features active in coding transitivity distinctions in sign languages. We trained linear support vector machines to make item-by-item transitivity predictions based on these features. Prediction accuracy was good across the entire dataset, thus demonstrating that systematicity in silent gesture can be explained with recourse to subunits. We argue that handshape features are constructs co-opted from cognitive systems subserving manual action production and comprehension for communicative purposes, which may integrate into the linguistic system of emerging sign languages. We further suggest that nonsigners tend to map event participants to each hand, a strategy found across genetically and geographically distinct sign languages, suggesting the strategy's cognitive foundation.

Gaps in the Lexicon Restrict Communication.

Across languages, words carve up the world of experience in different ways. For example, English lacks an equivalent to the Chinese superordinate noun tiáowèipǐn, which is loosely translated as "ingredients used to season food while cooking." Do such differences matter? A conventional label may offer a uniquely effective way of communicating. On the other hand, lexical gaps may be easily bridged by the compositional power of language. After all, most of the ideas we want to express do not map onto simple lexical forms. We conducted a referential Director/Matcher communication task with adult speakers of Chinese and English. Directors provided a clue that Matchers used to select words from a word grid. The three target words corresponded to a superordinate term (e.g., beverages) in either Chinese or English but not both. We found that Matchers were more accurate at choosing the target words when their language lexicalized the target category. This advantage was driven entirely by the Directors' use/non-use of the intended superordinate term. The presence of a conventional superordinate had no measurable effect on speakers' within- or between-category similarity ratings. These results show that the ability to rely on a conventional term is surprisingly important despite the flexibility languages offer to communicate about non-lexicalized categories.

Generative cultural learning in children and adults: the role of compositionality and generativity in cultural evolution.

Are human cultures distinctively cumulative because they are uniquely compositional? We addressed this question using a summative learning paradigm where participants saw different models build different tower elements, consisting of discrete actions and objects: stacking cubes (tower base) and linking squares (tower apex). These elements could be combined to form a tower that was optimal in terms of height and structural soundness. In addition to measuring copying fidelity, we explored whether children and adults (i) extended the knowledge demonstrated to additional tower elements and (ii) productively combined them. Results showed that children and adults copied observed demonstrations and applied them to novel exemplars. However, only adults in the imitation condition combined the two newly derived base and apex, relative to adults in a control group. Nonetheless, there were remarkable similarities between children's and adults' performance across measures. Composite measures capturing errors and overall generativity in children's and adults' performance produced few population by condition interactions. Results suggest that early in development, humans possess a suite of cognitive skills-compositionality and generativity-that transforms phylogenetically widespread social learning competencies into something that may be unique to our species, cultural learning; allowing human cultures to evolve towards greater complexity.

An information theoretic score for learning hierarchical concepts.

How do humans learn the regularities of their complex noisy world in a robust manner? There is ample evidence that much of this learning and development occurs in an unsupervised fashion via interactions with the environment. Both the structure of the world as well as the brain appear hierarchical in a number of ways, and structured hierarchical representations offer potential benefits for efficient learning and organization of knowledge, such as concepts (patterns) sharing parts (subpatterns), and for providing a foundation for symbolic computation and language. A major question arises: what drives the processes behind acquiring such hierarchical spatiotemporal concepts? We posit that the goal of advancing one's predictions is a major driver for learning such hierarchies and introduce an information-theoretic score that shows promise in guiding the processes, and, in particular, motivating the learner to build larger concepts. We have been exploring the challenges of building an integrated learning and developing system within the framework of prediction games, wherein concepts serve as (1) predictors, (2) targets of prediction, and (3) building blocks for future higher-level concepts. Our current implementation works on raw text: it begins at a low level, such as characters, which are the hardwired or primitive concepts, and grows its vocabulary of networked hierarchical concepts over time. Concepts are strings or n-grams in our current realization, but we hope to relax this limitation, e.g., to a larger subclass of finite automata. After an overview of the current system, we focus on the score, named CORE. CORE is based on comparing the prediction performance of the system with a simple baseline system that is limited to predicting with the primitives. CORE incorporates a tradeoff between how strongly a concept is predicted (or how well it fits its context, i.e., nearby predicted concepts) vs. how well it matches the (ground) "reality," i.e., the lowest level observations (the characters in the input episode). CORE is applicable to generative models such as probabilistic finite state machines (beyond strings). We highlight a few properties of CORE with examples. The learning is scalable and open-ended. For instance, thousands of concepts are learned after hundreds of thousands of episodes. We give examples of what is learned, and we also empirically compare with transformer neural networks and n-gram language models to situate the current implementation with respect to state-of-the-art and to further illustrate the similarities and differences with existing techniques. We touch on a variety of challenges and promising future directions in advancing the approach, in particular, the challenge of learning concepts with a more sophisticated structure.