The human ability to learn and compose conceptual operations is foundational to making flexible generalizations, such as creating new dishes from known cooking processes. Beyond naive chaining of functions, there is evidence from the linguistic literature that people can learn and apply context-sensitive, interactive rules, such that output production depends on context changes induced by different function orderings. Extending the investigation into the visual domain, we developed a function learning paradigm to explore the capacity of humans and neural network models in learning and reasoning with compositional functions under varied interaction conditions. Following brief training on individual functions, human participants were assessed on composing two learned functions, in ways covering four main interaction types, including instances in which the application of the first function creates or removes the context for applying the second function. Our findings indicate that humans can make zero-shot generalizations on novel visual function compositions across interaction conditions, demonstrating sensitivity to contextual changes. A comparison with a neural network model on the same task reveals that, through the meta-learning for compositionality (MLC) approach, a standard sequence-to-sequence Transformer can approximate a strong function learner, and also mimic human error patterns with additional fine-tuning.

With a series of theoretically-informed tests, Dankers, Bruni, and Hupkes (2021) investigated how compositional the behavior of neural networks that are trained on fully natural data is. Focusing on neural machine translation (NMT), one of their key findings is that models appear to be modulating poorly between local and global behavior, where local changes in the input often affect the output in an unwanted manner. While their study is based exclusively on the behavior of the models, we take one step further and investigate how this non-locality manifests itself within the model. We develop metrics to quantify internal locality on the encoder side of the model, focusing on the attention mechanism. We find strikingly different patterns in models trained on different amounts of data that go beyond what could be observed behaviourally and present a range of experiments showing how local and global behavior is modulated within different setups.

Language models have great potential as cognitive models for studying human language acquisition, but current models are far less data-efficient than human learners. Children acquire language from 100 million words or less, but large language models are trained on trillions of words. We discuss the prospects for improving language models' developmental plausibility through a meta-analysis of results from the 2023 BabyLM Challenge. BabyLM was a competition that invited participants to train a language model on a 100 million-word corpus including transcribed speech and child-appropriate texts. Results from over 30 submissions showed that new machine learning techniques and increased training iterations yielded models that outperformed leading large language models in grammar, language understanding, and linguistic generalization, while cognitively plausible approaches such as curriculum learning were less effective. We discuss the implications of these and other findings for computational cognitive modeling and explore ideas to ensure future competitions' contributions to cognitive science.